IMPROVING HEALTHCARE SUPPLY CHAINS AND DECISION MAKING IN THE

MANAGEMENT OF PHARMACEUTICALS

A Dissertation

Submitted to the Graduate Faculty of the

Louisiana State University and
Agricultural and Mechanical
in partial fulfillment of the
requirements for the degree of
Doctor of Philosophy
in
The Interdepartmental Program in Business Administration
(Information Systems and Decision Sciences)

by
John Michael Woosley
B.S., Louisiana State University, 1997
M.S., Louisiana State University, 2000
M.S., Louisiana State University, 2002
May 2009

DEDICATION

This work is dedicated to my family, especially

to my fianc, Amanda, for her love, support, patience, and friendship all of which make
me a stronger person and a better man;
to my parents, Barbara and Dwight, for teaching the value of hard work and for giving
me their unyielding support, encouragement, and love which helped me become
the man I am today and without which none of this would have been possible;
to my brother, Robert, for challenging me and helping me find my path in life;
to my niece, Madilyn, for being so very special and always reminding me what is truly
important; and
to Pogo and Tiger for reminding me when to stop working.
.

ii

ACKNOWLEDGEMENTS
Dr. Peter Kelle is a scholar and a gentleman, and I consider it an honor and a privilege to
have him serve as my advisor. He provided advice, guidance, and motivation necessary to keep
me focused and on my path. I will always be grateful for his commitment to helping me
complete of this work and my degree. More importantly, I am lucky to call him a friend.
Dr. Helmut Schneider has my sincerest gratitude for giving me the opportunity to join the
ISDS Department and to pursue this degree. I appreciate the confidence you have shown in me
over the years and will always value the experience I gained from working with you. I could not
have asked for a better mentor.
I want to acknowledge Dr. James Van Scotter, Dr. Sonja Wiley-Patton, and Dr. Boryung
Ju for their willingness to serve on my committee. Thank you for your time and support and for
helping me achieve my goal.
Dr. Suzanne Pawlowski has earned my deepest respect and admiration for being a
tremendous scholar and a wonderful person. I consider it a great honor to have you as both a
colleague and friend.
Finally, I would like to thank the remaining faculty and staff of the ISDS Department for
your kindness and friendship over the years.

iii

TABLE OF CONTENTS
DEDICATION ................................................................................................................................ ii
ACKNOWLEDGEMENTS ........................................................................................................... iii
LIST OF TABLES ......................................................................................................................... vi
LIST OF FIGURES ..................................................................................................................... viii
ABSTRACT................................................................................................................................... ix
CHAPTER 1: INTRODUCTION .................................................................................................. 1
1.1 Magnitude of Healthcare Industry and Impact of Pharmaceutical Costs ............................. 1
1.2 Special Conditions and Terminology of Healthcare SCM ................................................... 4
1.3 Traditional Operations in Hospital Pharmacy ...................................................................... 7
1.4 Research Objectives ............................................................................................................. 8
1.5 Organization of the Dissertation ........................................................................................ 10
CHAPTER 2: LITERATURE REVIEW ..................................................................................... 12
2.1 Managerial Research in Healthcare SCM .......................................................................... 13
2.1.1 Stakeholders and Relationships ................................................................................... 13
2.1.2 Value Chains................................................................................................................ 15
2.1.3 Stakeholder Conflicts .................................................................................................. 17
2.1.4 Managerial Approaches in Healthcare SCM ............................................................... 19
2.2 Quantitative Modeling in Healthcare SCM and Inventory Control ................................... 24
2.2.1 Inventory Control Policies and Planning Framework in Healthcare SCM ................. 25
2.2.2 Demand Specifics for Hospital Pharmacy ................................................................... 27
2.2.3 Multi-item, Single-location Models for Pharmacy Inventories ................................... 30
2.2.4 Single-item Multi-location Models - Multi-echelon Coordination in Pharmacy ........ 32
2.3 Practical Solutions for Inventory Management of Pharmaceuticals in Hospitals .............. 35
2.3.1 Current Management Trends and Operations in Pharmacy SCM ............................... 35
2.3.2 Multiple Criteria Decision Making (MCDM) ............................................................. 37
CHAPTER 3: RESEARCH ENVIRONMENT AND METHODS ............................................. 40
3.1 The Supply Chain and the Major Decisions of the Hospital Pharmacy ............................. 40
3.2 Quantitative Models for the Hospital Inventory Management........................................... 43
3.2.1 Model 1: A General Multi-product (s, S) Model with Space Constraint and Its
Approximation ....................................................................................................................... 43
3.2.2 Model 2: Optimal Allocation Based on Ordering and Holding Costs ........................ 47
3.2.3 Model 3: Optimal Allocation Based on Ordering Cost .............................................. 52
3.3 Comparison of Optimization Efforts for Model 2 and Model 3 ........................................ 55
3.3.1 Demonstration of Model 2........................................................................................... 55
3.3.2 Demonstration of Model 3........................................................................................... 61
3.4 Accuracy of the Approximations ....................................................................................... 63

iv

CHAPTER 4: ANALYSIS AND MANAGERIAL INTERPRETATION .................................. 66

4.1 Application for Managerial Decision Support ................................................................... 66
4.2 Levels of Decision Support ................................................................................................ 67
4.2.1 Operational Decision Support ..................................................................................... 68
4.2.2 Tactical Decision Support ........................................................................................... 74
4.2.3 Strategic Decision Support .......................................................................................... 82
4.5 Summary of Findings ......................................................................................................... 89
CHAPTER 5: CONCLUSION .................................................................................................... 90
5.1 Generalizability of Findings ............................................................................................... 90
5.2 Contributions ...................................................................................................................... 91
5.2.1 Contributions - Theoretical .......................................................................................... 91
5.2.2 Contributions Practical ............................................................................................. 92
5.3 Limitations ......................................................................................................................... 93
5.4 Future Research Directions ................................................................................................ 95
5.5 Concluding Remarks .......................................................................................................... 96
REFERENCES ............................................................................................................................. 98
APPENDIX A: PYXIS MEDSTATION AND MATRIX DRAWER ................................... 105
APPENDIX B: POWER APPROXIMATION .......................................................................... 107
APPENDIX C: MODEL 2 CALCULATIONS FOR REORDER POINT AND OPTIMAL Q 108
APPENDIX D: MODEL 3 CALCULATIONS FOR REORDER POINT AND OPTIMAL Q 109
APPENDIX E: MODEL 2 CALCULATIONS FOR SENSITIVITY ANALYSIS .................. 110
APPENDIX F: MODEL 3 CALCULATION FOR SENSITIVITY ANALYSIS ..................... 111
VITA ........................................................................................................................................... 112

LIST OF TABLES
Table 3-1: Changes in Overall Volume ....................................................................................... 56
Table 3-2: Values of Lambda ...................................................................................................... 57
Table 3-3: Average Percent Change in Q with Iteration.............................................................. 57
Table 3-4: Optimization of Q for a Sample Drug ........................................................................ 59
Table 3-5: Functions of Q for Model 2 Iterations ........................................................................ 59
Table 3-6: Values of M for Iterative Rounds ............................................................................... 60
Table 3-7: Order Quantities and Reorder Points for Sample ....................................................... 60
Table 3-8: Fixed-days Supply Policy of Hospital ........................................................................ 61
Table 3-9: Inputs for Model 3 ...................................................................................................... 62
Table 3-10: Percent Change in Space Available to Cycle Stock (M) .......................................... 62
Table 3-11: Optimal Values from Model 3.................................................................................. 63
Table 3-12: Accuracy of the Approximations - Comparison of the Calculated and Simulated Key
Performance Indicators ................................................................................................................. 64
Table 4-1: Percent Differences in Model Costs to Hospital Policy ............................................. 70
Table 4-2: Cost Comparison of Model 2 and Model 3 ................................................................ 70
Table 4-3: Policy Comparison with Constant Space Utilization (M) .......................................... 75
Table 4-4: Policy Comparison with Constant Service Level (SL) .............................................. 75
Table 4-5: Policy Comparison with Constant Service Level and Increased Space ..................... 76
Table 4-6: Policy Comparison with Constant Number of Daily Refills (N) ............................... 77
Table 4-7: Comparison of the Different Space Allocation .......................................................... 77
Table 4-8: Comparison of Space Available for Cycle Stock (V) at Varying Holding Costs ...... 78
Table 4-9: Percent Change of Space Available for Cycle Stock (V) at Varying Holding Costs . 79
Table 4-10: Impact of Changing Holding Costs on Available Space (V) ................................... 79
vi

Table 4-11: Optimal Order Quantities (Qi) at Differing Holding Costs ...................................... 80
Table 4-12: Pairwise Comparisons of Percent Differences for Varying Holding Cost ............... 81
Table 4-13: Impact of Adding or Subtracting Items to Local Formulary .................................... 87
Table 4-14: Incremental Increases to Worker Capacity .............................................................. 88

vii

LIST OF FIGURES
Figure 2-1: Stakeholders in the Healthcare Value Chain (Pitta and Laric 2004) ........................ 16
Figure 3-1: Convergence to the Optimal Q for a Sample Drug ................................................... 58
Figure 3-2: Approximation of Target Service Level ................................................................... 65
Figure 4-1: Cost Comparison for Allocation Methods ................................................................ 69
Figure 4-2: Performance Indicator Relationships and Tradeoffs ................................................. 85

viii

ABSTRACT
The rising cost of quality healthcare is becoming an increasing concern. A significant
part of healthcare cost is the pharmaceutical supply component. Improving healthcare supply
chains is critical not only because of the financial magnitude but also because of the fact that it
impacts so many people. Efforts such as this project are essential in understanding the current
operations of healthcare pharmacy systems and in offering decision support tools to managers
struggling to make the best use of organizational resources.
The purpose of this study is to address the objectives of a local case hospital that are
typical general problems in pharmacy supply chain management. We analyze the pharmacy
supply network structure and the different, often conflicting goals in the decisions of the various
stakeholders. We develop quantitative models useful in optimizing supply chain management
and inventory management practices. We provide decision support tools that improve
operational, tactical, and strategic decision making in the pharmacy supply chain and inventory
management of pharmaceuticals.
On one hand, there is considerable progress to support pharmacy product distribution by
advanced computerized technology that manages the dispensation of medications, and automates
the ordering. On the other hand, the available information is not utilized to help the managers in
making the appropriate decisions and control the supply chain management.
Quantitative methods are presented that based on the available data provide simplified,
practical solutions to pharmacy objectives and serve as decision support tools. For the
operational inventory decision we provide the min and max par levels (reorder point and order
up to level) that control the automated ordering system for pharmaceuticals. These parameters
are based on two near-optimal allocation policies of cycle stock and safety stock under storage
space constraint. For the tactical decision we demonstrate the influence of varying inventory
ix

holding cost rates on setting the optimal reorder point and order quantity for items. We present
a strategic decision support tool to analyze the tradeoffs among the refill workload, the
emergency workload, and the variety of drugs offered. We reveal the relationship of these
tradeoffs to the three key performance indicators at a local care unit: the expected number of
daily refills, the service level, and the storage space utilization.

CHAPTER 1: INTRODUCTION
A recent visit to a local emergency room (ER) motivated the author of this work to
literally investigate the crucial element of supply chain management (SCM) within the healthcare
system. After experiencing eight hours of delays in the ER waiting room and witnessing
incredible lags in the medication dispensing process and even further impediments to the
delivery of treatment modalities, the researcher became outraged with the thought of a highperforming healthcare systems lack of better supply chain management and service delivery. As
a healthcare consumer, this scenario is likely to happen to all of us.
One might contemplate how an industry based on customer service falls short in the arena
of inventory, patient and materials management? According to business researcher Vicki SmithDaniels (2006) in a field where precision is literally a matter of life and death, it seems strange,
that a crucial supportive function like inventory control and purchasing is often a hit-or-miss
process. Healthcare, unlike other industries has not given supply chain management the detailed
attention that it so rightly deserves and needs to ensure patient safety and reduce overall
healthcare costs.
Inventory management in the healthcare industry presents several interesting challenges
both from a managerial and operational perspective. The stakeholder relationships, product
considerations, and managerial and regulatory policies typically seen in healthcare are unique
and worth investigation. Before delving into specific inventory control issues or demand
characteristics that make healthcare supply chain management (SCM) particularly difficult, it is
important to recognize the magnitude of this industry.
1.1 Magnitude of Healthcare Industry and Impact of Pharmaceutical Costs
According to statistics published by the Centers for Medicare and Medicaid Services,
healthcare spending topped $2 trillion or 16% of its Gross Domestic Product (GDP) in the
1

United States in 2005 (Centers for Medicare and Medicaid Services 2007). In addition, this
percentage is projected to increase to 18.7% in ten years (Heffler et al. 2005). Catlin et al. (2007)
state that healthcare expenditures increased at an annual rate of 6.9%. In addition, there has been
a shift from public to private healthcare financing in the United States, as well as other countries
around the world. Most of these efforts are an attempt to continuously find new ways of both
controlling healthcare expenditures and paying for prescribed treatments and pharmaceuticals.
Especially given the current economic crisis, a growing amount of attention is being
given to the rising costs of healthcare and specifically pharmaceuticals. Healthcare providers,
insurance companies, government agencies, and consumers alike are forced to address this issue
and to explore alternative methods of cost reduction or cost containment (Marmor and Okma
1998; Jnsson and Musgrove 1997; Culyer 1990). To gain a better understanding of the
significance of this issue, it is prudent to first identify the magnitude of healthcare expenditures.
According to The Plunkett Research Group (2008), The health care market in the U.S. in 2007
was made up of hospital care (about $697.5 billion), physician and clinical services ($474.2
billion), prescription drugs ($229.5 billion), nursing home and home health ($190.0 billion), and
other items totaling $668.8 billion.
As shown above, the financial aspect of the healthcare and pharmaceutical industries is
quite staggering. Much of the research in this field has focused on the relevant costs of
healthcare (Castles 2004; Comas-Herrera 1999; OECD 1994; OECD 1995; OECD 1996);
however, Rothgang et al. (2005) examined data reported by the Organization for Economic Cooperation and Development (OECD) from 1970 to 2002 to investigate trends in the level of
governmental involvement with healthcare systems. Although some shifting has occurred over
the years, convergence towards a mixed system of both public and private healthcare seems
apparent.
2

A significant area of healthcare costs is the pharmaceutical area, which represents

approximately ten percent of annual healthcare expenditures in the United States and about $550
billion globally in 2007 (Plunkett Research 2008). Despite the size and importance of this
industry around the world, especially in developed countries, the area of healthcare supply chain
management (SCM) and inventory management has been given relatively little attention.
Several researchers have estimated that inventory investments in healthcare range between 10%
and 18% of total revenues (Holmgren and Wentz 1982; Jarrett 1998). Any measures taken to
control expenditures in this area can have substantial impacts on the overall efficiency of the
organization and its supply network and, as a result, the profitability of healthcare providers.
In 2003, Guillen and Cabiedes examined the pharmaceutical policies of European Union
(E.U.) countries from the mid-1980s through the 1990s. As explained in their research, the costs
of pharmaceuticals continue to increase while countries struggle to find financing to support
these drugs. They noted that much of the healthcare services are shaped by the pharmaceutical
policies, and they also identify three reasons justifying a focus on pharmaceuticals when
examining healthcare supply chains: the reliance of modern medicine on drugs to both prevent
and cure sickness; the significance of pharmaceutical expenditures around the world; and the
pharmaceutical industry characteristics (i.e. use of technology, innovation, etc.). Almarsdttir
and Traulsen (2005) also identify a number of reasons why pharmaceuticals deserve
extraordinary consideration in controlling inventory. These specific characteristics make the
pharmaceutical industry a very powerful force in its own right.
As with any industry the range of products can vary tremendously depending on the
market, customer demand, the scope of services, and other managerial decisions. For the
purposes of this project and the specific research case presented, the focus is mainly on the
supply chain of pharmaceuticals within the hospital.
3

1.2 Special Conditions and Terminology of Healthcare SCM

The stakeholders and their interrelationships, the product characteristics, and the policies
employed all have some impact on the healthcare supply chain management and inventory
control. An explanation of some specific conditions and terminology used in healthcare are
discussed in more detail below.

Stakeholders
Here is a brief summary of some unique factors influencing how healthcare stakeholders

interact, how they are controlled, and how they manage operations. A detailed analysis is
provided in the next section.
Doctors are the primary caregivers in the healthcare system; however, it is important to
recognize that physicians, in many cases, are contracted service providers. They are not actual
employees of hospitals in which they work. Although this enables hospitals to expand their
service capacities and offer better customer service, they must also relinquish some measure of
control to these doctors. Autonomy, as it relates specifically to customer care, is valued above all
else by doctors. Patients trust that physicians are prescribing treatments and medicines that will
address their individual medical needs. Attempts to restrict the choices or influence the
conditions of medical treatment meet with a great deal of resistance.
Another consideration is that, unlike many industries of this size, hospital
administrators and pharmacy managers have to manage very complicated distribution networks
and inventory control problems without the proper training or educational backgrounds to do so
efficiently. This in no way implies that these individuals lack the intelligence to perform these
tasks. Most hospital administrators and pharmacy directors are themselves doctors, which means
they are highly skilled and educated in medicine; however, they are not engineers or supply
chain professionals that are commonly employed to manage similar systems in other industries.
4

The hospitals order the majority of the supply from a selected group purchasing
organization (GPO) that purchases directly from the production companies.

Products
The product formulary is the term used to identify the variety of drugs offered by the

hospital pharmacy and is a source of both conflict and cost for the hospital. This formulary is
comprised of specific medicines each designed to address a particular medical need, but
physicians opinions may vary on the most effective drugs at satisfying patient requirements.
Providing prescription options for the doctors is important, both to the doctors and hospital;
however, this increases the number of drugs that must be carried by the hospital pharmacy for
patient treatment if it desires to maintain such a high level of customer service. In addition, the
product formulary changes quite frequently as physicians prescribing behavior reacts to
advancements in medical research and technology.
It is also important to recognize that pharmaceuticals have a number of other
requirements. They must be handled by trained personnel and experts in this field.
Extraordinary resources are committed to developing these items, to manufacturing them, and to
controlling their distribution and usage. All of these aspects are highly regulated by
governments and other regulatory agencies, and they may require additional documentation.
Finally, unlike other areas, customers have a limited role in the product selection process. As
patients become more aware of alternative treatments, they may influence doctors as they
prescribe medications; however, more times than not these choices are made for them.
Many of these items are also perishable and must be destroyed if not used. Perishable
items have a limited shelf life and in many cases have special transportation and storage
requirements. Part of the concern with pharmaceuticals is that outdated or expired items may be
overlooked and dispensed to patients, which could have potentially disastrous effects both in
5

patient care and public relations. The perishable item inventory control problem is a difficult one
and has been studied many times using a periodic review approach (see Fries 1975, Nahmias
1975, Nahmias 1982) with less attention being given to continuous review systems; however,
there have been a number of works in this area in recent years (see Weiss 1980; Schmidt and
Nahmias 1985; Chiu 1995; Liu and Lian 1999; Lian and Liu 2001). Although these studies
provide valuable insight into the management of such products, none of these models have been
tested in pharmaceutical inventory control. Given the combination of high costs and
perishability of prescription drugs, it seems that more study is warranted as pharmacy managers
look for help in setting optimal order policies.

Policies
Substantial efforts are made to regulate the healthcare industry. Government

involvement is high as it can provide both oversight of caregivers and funding for care recipients.
Previously, healthcare systems have paid little attention to the management of inventories.
However, with the implementation of Diagnostic-Related Groups (DRG) in 1983 by the United
States government, these systems have turned their attention to cost containment as a means of
increased profitability. The original objective of diagnosis-related groupings (DRGs) was to
develop a patient classification system that related types of patients treated to the resources they
consumed. Hospital cases are segmented into one of approximately 500 groups expected to have
similar hospital resource use. DRGs are used to determine how much Medicare pays the
hospital, since patients within each category are similar clinically and are expected to use the
same level of hospital resources.
Interestingly, public and managerial policies targeted at cost-control have failed in many
cases to produce the necessary changes in stakeholder behaviors to achieve such outcomes.
Efforts to control pharmaceutical prices have provided little relief as drug manufacturers attempt
6

to recover the resources spent researching and developing these products. These companies also
target physicians and patients with advertising campaigns as they attempt to influence
prescribing behavior. Another reason policy has been somewhat ineffective at cost-restraint is
the conflicting stakeholder goals that appear throughout the supply chain. Doctors and hospital
administration are often at odds as they try to balance the issues of prescribing autonomy and
product variety. While the doctors want a large variety of brand name drugs, the hospital
management strives to minimize costs in the overall system by seeking generic drugs as
substitutes for medicines preferred by physicians. Hospital and GPO have different objectives.
Hospitals focus on negotiating the best prices for a wider selection of drugs from different
pharmaceutical companies while the GPOs are interested in larger orders from only a few
pharmaceutical production companies.
Inventory control variables are called par levels in healthcare inventory management.
The min par level is equivalent to the reorder point: if the inventory level decreases to or below
this level, an order is triggered. The max par level is equivalent to the order up level (or base
stock).
1.3 Traditional Operations in Hospital Pharmacy
Traditionally, hospital pharmacies operated in a centralized manner. Medications were
stored in a central pharmacy and then distributed by pharmacy technicians using dose carts to
make deliveries to the various hospital floors and Care Units (CUs). A setup such as this
requires large amounts of inventory to be stored in a single location while also allowing for
quick, easy access to items. These technicians were charged with the task of moving the various
medicines in the correct amounts or in individual cassettes for each patient. These dose carts
contained individual cassettes of medications for each patient, which the nurse or caregiver was
then charged with administering to the patient. If for some reason the patients drug order
7

changed, a special trip would be required to retrieve the necessary medications and then return to
the CU. Thus, changes in medication or dosage may have taken hours to satisfy resulting in time
delays in care, increased labor requirements, and an overall increase in system costs associated
with wasteful or excess operations.
To avoid prolonged delays in administering medications, it was common practice for
nurses to borrow a medication from another patient. That resulted in additional problems in the
distribution process and in potentially dangerous situations where patients may have received the
improper medications. Other difficulties associated with the specific pharmaceutical types
existed. Specifically, federal and state regulations require a variety of documentation associated
with the dispensation of narcotic drugs. From the time a narcotic is delivered to the hospital until
it is administered to the patient, the drug must be tracked by the hospital and is surrounded by a
number of tasks such as checking the ordered medication against the patients medical record,
searching for narcotic keys to the cabinet, documenting for administrative records and
reconciling narcotic records after each shift. This example demonstrates some specific
challenges in controlling and monitoring items in healthcare SCM.
Pharmacists are highly paid professionals being asked to spend long amounts of time
handling excessive documentation and a labor intensive drug dispensation and distribution
process rather than practicing pharmaceutical care. The opportunity to reduce repetitive and
tiresome tasks was very appealing to administrators and pharmacists alike. As such, the ability
to automate these processes and take advantage of computer information systems was valued by
stakeholders.
1.4 Research Objectives
The hospital participating in this investigation employs an inventory management
solution driven by information technology (IT). The specific solution and operating procedures
8

are discussed in more detail in Chapter 3; however, it is prudent to provide a simple overview of
this system at this point. Specifically, this solution, Pyxis MedStation allows for local storage
depots to be distributed at the various CUs around the hospital. These Pyxis machines house
the drugs needed for patient care in that CU, as well as, track every inventory transaction, prompt
replenishment orders, generate necessary documentation, and facilitate the billing processes
related to pharmaceutical treatments. While these Pyxis MedStations allow for the automation
of a number of tasks associated with this supply network, it is important to recognize that
pharmaceutical inventory management is still a very labor-intensive process due to the number
of Pyxis machines in the hospital (~85), the large volume of drugs housed in each depot (250500), and the workload required during the restocking process. As such, the following research
objectives are identified:
1. Service Leveli max, > i
Here we are trying to maximize or set a lower limit for service levels for each
item in the Pyxis machine. The service level is defined as an appropriate measure of
customer service for each of the n items (i.e. 90%, 95%, or 99%). Alpha () service
levels are associated with items based on the number of shortage occasions that managers
are willing to accept during a period of time. If for some reason the hospital pharmacy is
out of this item, it can place an emergency order with its supplier to replenish this item.
However, the hospital would like to avoid these emergency orders if possible because
they are very costly for the organization. For this objective we are considering the
service level, , for each item separately. To consider all items in the machine as a whole
would result with lesser demanded items being unavailable more often to accommodate
the higher demanded goods. In addition, these levels reflect the average service level for
the item rather than the worst-case scenario level as set by the hospital.
9

2. i (Orders per Dayi ) min, N(workload constraint)

Here we are trying to minimize the number of orders per day (refill occasions) or
at least keep this number less than the number of orders a worker can handle during a
shift. This is an attempt to account for the workload constraint. Workers can perform a
limited number of restocking activities during a given shift. There is an incremental
workload constraint, which only increases with the addition of trained pharmaceutical
technicians. As such, the goal here is to find inventory control values that will minimize
the number of required refill occasions per day and not exceed the work capacity for a
single worker.
3. i (vi Si ) M min (space constraint)
Here the goal is to allocate space for each item in the Pyxis unit such that the
unit contains the necessary drugs but does not exceed the available space. The goal is to
determine the appropriate amount of space necessary for each item within a Pyxis unit
by first assigning space to accommodate si (safety stock) and then determine the
additional space required to house Di, which is the difference between the order-up-tolevel (Si) and the minimum stock level (si). This is done to provide support for
reorganizing dividers within the Pyxis unit to allocate the available space appropriately
amongst all of the required items.
1.5 Organization of the Dissertation
The remainder of this document is divided into several chapters with each addressing
specific topics related to this research. Chapter 2 offers a review of relevant managerial and
quantitative modeling literature, as well as, presents practical solutions employed to address
challenges specific to hospital inventory control. Chapter 3 presents the research environment
and the modeling approaches used to satisfy the investigation objectives. Chapter 4 provides the
analysis and findings of this study. In addition, the impacts of these results are examined for
10

managerial decision support purposes. Chapter 5 brings the conclusion of this dissertation, as
well as, promotes areas of future study.

11

CHAPTER 2: LITERATURE REVIEW

This research focuses specifically on supply chain management issues from the
perspective of the hospital. Other members of the supply network in this industry are
important; however, the subject is simply too voluminous to cover adequately in this setting
without restricting the scope somewhat. As such, the authors have selected to focus on supply
chain problems as they are viewed by hospitals, which is appropriate given their significance in
the supply chain and the costs they incur. In the following sections critical managerial areas are
identified that distinguish the healthcare environment from other industries that have received
attention in the past.
Pharmacy directors are turning their attention to controlling costs in the areas of supply
chain management and pharmaceutical inventory control. Supply and purchased services
account for the second largest cost component for a hospital, and it is clearly recognized that
SCM is one of the principal areas for improvement in organizational performance. Supply chain
management in hospital systems all over the world shows a great variability both in performance
and in focus. Such a variability combined with the growing cost of care and pharmaceuticals
determines relevant differences in health service efficiency and unacceptable annual increases in
healthcare costs.
The remainder of this chapter will serve to address several key areas of healthcare SCM
and pharmaceutical inventory management. A review of the relevant managerial and
quantitative modeling literature is provided. Here attention will be given to both the overall
study of healthcare SCM and the quantitative approaches to inventory optimization. The practice
of pharmacy operations will be described with specific attention given to the movement and
distribution of prescribed medications within a hospital. Solutions focused on addressing SCM

12

challenges will be presented along with the Multiple Criteria Decision Making (MCDM) nature
of decision making in this context.
2.1 Managerial Research in Healthcare SCM
Traditionally researchers have taken two approaches to studying healthcare SCM:
analyzing the managerial concerns and evaluating various quantitative models. First, the
managerial aspects of such supply chain systems are discussed. The stakeholders in healthcare
supply chain and value chain are examined along with the conflicting goals in the system.
Thereafter the most common managerial approaches in healthcare SCM are summarized
including outsourcing and vendor managed inventory (VMI). In existing research, the focal point
has primarily been on description of the system rather than developing innovative models or
performing quantitative analyses. However, these descriptive studies are important to document
the actual healthcare supply practices.
2.1.1 Stakeholders and Relationships
The financial aspects of the healthcare industry are significant, but it is also an industry
very important to all members of society as it involves a wide range of stakeholders. Burns
(2002) examined the healthcare industry for three years to investigate its value chain, to uncover
significant industry trends, and to identify the major stakeholder groups involved with healthcare
services. Evidence showed that both vertical and horizontal integration were present in
healthcare. Vertical integration was illustrated by hospitals teaming with insurance agencies or
ambulatory services to combine various portions of this delivery network. Noticeable horizontal
integration manifested in the form of hospitals purchasing other hospitals or in the formation of
groups purchasing organizations. A significant result of this research was the identification of
the following groups as stakeholders in this industry: 1) payers, 2) fiscal intermediaries, 3)

13

providers, 4) purchasers, and 5) producers. Each of these groups is explained in more detail
below.
Any person or organization responsible for supplying the funds to pay for medical
expenses is identified as a payer. Based on this definition, examples of payers would include
government, employers, individuals and employer coalitions. Insurance agencies, health
maintenance organizations (HMOs), and pharmacy benefit managers fall under the category of
financial intermediaries. Any group supplying healthcare screening, treatment, or any other
healthcare related provisions are considered to be providers. Such entities would be physician
offices, hospitals (or hospital systems), surgical centers, alternative and satellite facilities,
ambulatory services, and pharmacies. Individuals or groups procuring any of these services are
viewed as purchasers; however, it is important to note that this category extends beyond the
basic consumers of healthcare services. This group also includes a wide variety of product
resellers, independent distributors, pharmaceutical wholesalers, etc. The final group is the
producers whose responsibility is the manufacturing of healthcare products, equipment, and
technology. This includes pharmaceuticals, surgical equipment, information technology
services, medical devices, and other capital equipment found throughout the healthcare system.
Tarabusi and Vickery (1998) state that attitudes and habits of local pharmacists and
physicians must be known for good access to markets. This research identifies the importance of
cost-containment programs in the pharmaceutical industry by comparing the approaches
employed by the United States and countries of the European Union. Globalization and
international partnerships have grown as pharmaceutical companies strive to control the research
and development costs associated with these drugs.

14

2.1.2 Value Chains

The notion of value in the healthcare service value chain can be described by the
interrelations of relevant stakeholders. Value chains were first introduced by Porter (1985) and
are significant marketing tools because they afford managers the opportunity to assess the
specific value added by each member of the supply network. Supply chains provide a
mechanism for delivering that value to consumers. The original view of a value chain focused
on a companys internal activities specifically designed to create value for customers; however,
this concept has been extended to include the entire product and service delivery system (Bower
and Garda 1985; Evans and Berman 2001).
Pitta and Laric (2004) provide a model of the healthcare value chain and supply chains as
they exist in many practical situations. Figure 2-1 below illustrates that this supply chain is not
linear or sequential in nature but closely follows the flow of information through the system.
The success or value created is linked to the transfer of quality information as the medical care
received by patients relies heavily on information processing.
The first two groups interacting in this network are the patients and physicians. This
stage of the process is generally initiated by the patient and provides valuable information
necessary to adequately address whatever needs they might have. This research showed that
individuals are much more likely to share very personal information with their healthcare
providers when they believe this information is needed for medical purposes and when they trust
the confidential nature of the patient-doctor relationship. The next link in the chain is created by
the addition of pharmacists and other providers of medical equipment and services. In this stage
the pharmacist creates value by further investigating the medical history, specifically as it relates
to medications, of patients to determine any potential risks or interactions that may result from

15

the addition of new, prescribed drugs. This is a very important step in customer service and
patient care.

Figure 2-1: Stakeholders in the Healthcare Value Chain (Pitta and Laric 2004)
The next addition to the value chain is that of the hospital and the related services and
procedures that may be included in diagnosing patient symptoms. As explained by Pitta and
Laric (2004), doctors often request batteries of tests be performed on patients even when
symptoms and other indicators suggest a course of treatment. Hospitals create and store vast
amounts of medical data for patients, which can be useful going forward. The fifth member of
the healthcare value chain is the health insurers, which includes both public and private entities
responsible for providing financial support to those receiving care. Since many insurers require
a series of diagnostic tests and related data be completed before approving potentially costly
procedures, patient data continues to grow. The addition of insuring groups or companies can
also be a negative influence on the value chain. As one can see with the addition of new
16

members to the network and the creation of data at each stage, the healthcare system and
participant interactions are becoming more and more complex.
Employers are the sixth group added to create another value chain. In the United States
(US) employers are the most often used source of medical insurance for employed persons and
their families. Here the value is obtained by negotiating better benefits for groups of people as
opposed to those offered to individuals. When the US Government introduced its Federal
Medicare and Medicaid programs in the 1960s, it became the largest medical insurance provider
in the United States. A result of this involvement is the increasing influence of government
regulation and policy in the healthcare industry. As such, government becomes the seventh
participant of interest here as it influences various parts of this system. The final member of the
model is the pharmaceutical manufacturers. It has already been established pharmaceuticals
represent a significant area of cost in the healthcare industry, as well as being the primary means
of preventative and curative medical treatment.
2.1.3 Stakeholder Conflicts
Physicians and pharmacists/pharmacy directors clash over medications offered by the
hospital. The basic conflict here revolves around the issue of product variety versus economies
of scale. The doctors have professional and personal preferences that are reflected in their
prescription decisions. They value their individual freedom of choice in selecting the
medications that they feel best address the specific needs of the patients under their care.
Physicians are also influenced by drug manufacturers, sales representatives, and the appearance
of new drugs in the market. In contrast, the pharmacy directors pay more attention to the costs
associated with specific medications and promote the usage of generic rather than brand drugs
when available. They strive to take advantage of economies of scale whenever possible in the
selection of drugs offered.
17

In 2005, Prosser and Walley examined the extent to which cost influenced prescribing
behaviors of general practitioners (GPs) in the UK healthcare system. According to the authors,
Primary Care Organizations (PCO) involvement has steadily increased as a means of providing
more unified, controlled budgets for these physicians and to provide a monitoring mechanism of
prescribing activities with this cost-containment objective in mind. The primary goal of this
investigation was to measure the cost-awareness of doctors and to evaluate the attitudes of these
GPs towards cost-restraint policies. Another goal was to determine what affect this managerial
objective might have on their professional behavior. In this case the evidence demonstrated that
all of the physicians were aware of pharmaceutical costs when prescribing treatments for their
patients and that these costs should be considered when making prescription choices. However,
there appeared to be a great deal of variation in the amount of influence these costs actually had
on modifying medical decisions. As expected, physicians identified patient care as their primary
focus during the treatment process with cost being a secondary concern.
Prosser and Walley (2007) continued their research in this area and employed qualitative
methods to examine the managerial influence of healthcare administrators on the prescribing
autonomy of general care practitioners. In general, physician prescribing autonomy has been
challenged recently due to the greater sophistication of patients in the ability to direct medical
decisions and the increased bureaucratic involvement in service delivery in healthcare. This
research showed that the objective of cost-containment presents a formidable obstacle for
caregivers striving to offer the highest level of care and individualized service to those they
serve.
The above analysis showed that managerial involvement will have limited success in
influencing GPs prescribing autonomy. The conclusion was that the GPs directly resisted the
attempts administrators made to modify their prescribing behaviors because the GPs maintain it
18

is their responsibility to offer the best medications and treatment available to their patients and
that this autonomy is a component of providing individual care. In many cases this is an area of
great conflict between doctors and hospital/pharmacy administrators since larger product
formularies desired by doctors usually result in higher inventory costs and reduce the opportunity
for cost savings. When asked, doctors acknowledged the importance of controlling the costs of
pharmaceuticals, but they identified the quality of patient care as being the paramount objective.
Managers placed a great deal of importance on keeping prescribing costs within established
budgets.
Another divergence appears between hospitals and the group purchasing organization
(GPO) on the issue of product variety. Hospitals focus on negotiating the best prices for a wider
selection of drugs. The attention is shown to the brand name medications in most cases. The
GPO, on the other hand, strives to minimize costs in the overall system by seeking generic drugs
as substitutes for medicines preferred by physicians. Again, a great deal of effort is given to the
achievement of economies of scale for demanded items by considering a limited formulary
(narrower selection of drugs).
The importance of product standardization for managing costs and improving clinical
performance is outlined with respect to physicians unwillingness to accept products alternative
to branded ones, even in the face of evidence regarding equivalence. As such, hospitals persist
in the argument for wider drug selections. Both physician and nurse involvement and leadership
in product analysis is necessary for successful standardization to occur and for the provision of
metrics pertaining to products and their relationship to the performance of the supply chain.
2.1.4 Managerial Approaches in Healthcare SCM
In this section of the chapter, the focus shifts to the various strategic approaches that have
been pursued in the area of supply chain management and inventory control. Some of the topics
19

to be discussed are the outsourcing of distribution activities, allowing suppliers to manage

inventory levels at various distribution points, and the use of common statistical techniques to
achieve organizational and system goals.
2.1.4.1 Outsourcing
Kim (2005) presents an explanation of an integrated supply chain management system
developed to specifically address issues related to pharmaceuticals in the healthcare sector.
Many industries have recognized the importance of improved information sharing throughout the
supply chain, and this work considers it as the most critical success factor. Here the supply
network is composed of pharmaceutical companies, a wholesaler, and hospitals. The hospitals
operating procedures and policies were reviewed to determine system requirements in an effort
to improve the management efficiencies of the supply chain.
Jayaraman et al. (2000) present several tools and practical ideas to improve the flow of
materials in a small healthcare facility. Traditional techniques, such as Pareto diagrams and
department-product-type (DPT) matrices, were employed to track item flows and identify
sources of errors or difficulties. Researchers suggested several procedural and policy changes be
implemented to reduce inventory management problems. Landry and Beaulieu (2000) present a
descriptive study of logistics systems at hospitals from three countries France, Netherlands and
The United States to identify the best practices for replenishment policies, equipments, and
handling technologies.
Nicholson et al. (2004) compared the inventory costs and service levels of an internallymanaged three-echelon distribution network and contrasted it with an outsourced two-echelon
distribution network. Their research revealed that a general trend in healthcare is the outsourcing
of specific organizational activities, inventory management, and materials distribution to
expert third-party providers. To this point very little attention has been given to inventory
20

management; however, changes in pharmaceutical regulations and the formation of the DRGs
prompted a shift in approach.
When outside entities can provide needed products or services more efficiently than
internal departments, it can prove very rewarding. In other instances outside providers may
demonstrate the ability to provide the desired products or services at a higher level of quality
than an organization may be capable of achieving (Lunn 2000). A long-term benefit of
outsourcing is the ability to reduce the number of suppliers in the system, which will eventually
lower the procurement costs for the downstream members of the supply chain. As such, the
short-term benefits of increased efficiency and higher quality along with the long-term benefit of
lower procurement costs make outsourcing a logical approach to overall cost containment in the
healthcare industry and consistent with current practices in inventory management (Veral and
Rosen 2001).
Outsourcing decisions in this area are motivated by three factors: the magnitude of
investment, the impact on service quality and delivery, and the availability of qualified service
providers. Quality of service is arguably the paramount goal of healthcare providers (Li and
Benton 1996), and it has been suggested that outsourcing various functions allows organizations
to improve the quality of its internal operations. Jarrett (1998) identified several areas in which
providers were able to improve internal performance. Patient care was improved, which can
influence customer satisfaction and perceptions related to the quality of service provided by an
organization. As the expertise develops and the capabilities of external sources grow,
outsourcing becomes an increasingly attractive option. Another motivation for healthcare
providers considering outsourcing the inventory management functions are the success stories
similar to those described by Rivard-Royer et al. (2002).

21

Rivard-Royer et al. (2002) examined the changing role of distributors in the healthcare
industry in recent years by investigating how operational processes had changed at a Quebec
hospital As reported by Jarrett (1998), the ever-increasing desire to control healthcare costs have
led hospitals to re-evaluate many of their previous policies and operations. Inventory control is a
logical area where substantial gains can be realized. Pharmaceuticals represent a significant cost
equation related to both hospital inventory and quality patient care. As such, many hospitals and
hospital systems have focused on this specific material management issue when restraining costs.
US pharmaceutical distributors were offering stockless replenishment where the distributors
would prepackage items based on usage at individual care units and would often provide direct
delivery of drugs to these CUs (Henning 1980). By employing such a system the distributors
share in the benefits and risks associated with pharmaceutical savings or costs. The reduction in
pharmaceutical inventory located at the hospital central pharmacy saved hospitals valuable
resources, and the shifting of duties from the pharmacies to the distributors allowed the hospitals
to reduce their workforce. Another improvement was noticed in enhanced customer service.
However, as one would expect, a critical success factor in achieving any benefits from such a
relationship and process is the continuous exchange of information between the point of use and
supplier (Bolton and Gordon 1991).
2.1.4.2 Supply Coordination and Vendor Managed Inventory
According to Rivard-Royer and colleagues (2002), by the late 1990s stockless
replenishment was a thing of the past as hospitals sought a balance between the amount of effort
being spent in replenishing hospital inventories and the hospitals inventory savings. The
hospital involved with this study employed a hybrid approach since high-demand items were
purchased in bulk (at the case level) for delivery directly to the patient CUs from distributors and
low-volume items were handled by the pharmacy Central Storage (CS). For these low-demand
22

items the CS was charged with breaking-down the bulk purchases into smaller quantities for
distribution to the various CUs as items were used. However, this hybrid approach was found to
have limited benefits to both the hospital and distributors. Distributors were asked to perform
additional work without receiving a comparable level of additional revenue. The hospital was
still unpacking, repacking, and preparing drugs for distribution to the CUs and failed to realize
significant workload reductions as a result of this replenishment approach. Overall, the gains
were very limited in this particular study.
One managerial shift has been to vendor managed inventory (VMI), which has been
shown to have benefits in SCM related to enhanced material handling efficiency. Other trends in
supply coordination have included online procurement systems and the availability of real-time
information sharing. These advances have demonstrated an ability to reduce total
pharmaceutical inventory by more than 30% (Kim 2005). In addition, the improved information
sharing throughout the supply chain allowed for more timely and accurate inventory data, which
resulted in better demand forecasts and materials management.
The research of Meijboom and Obel (2007) investigated supply chain coordination in a
global pharmaceutical organization. The authors looked at the issues facing a pharmaceutical
company with a multi-location and multi-stage operations structure. Simulations are used to
explore various coordination activities focusing specifically on tactical control in the firm.
Their results suggest that a functionally organized operations structure will be unsuccessful in a
multi-location, multi-stage environment. According to the authors, firms can employ one of two
methods. They can either use a centralized approach relying heavily on information systems or
maintain a decentralized structure that relies on transfer prices as the coordination device.
Lapierre and Ruiz (2007) developed two modeling approaches for improving healthcare
inventory management by examining the impact of scheduling decisions on the coordination of
23

supply activities while recognizing inventory capacities. Specifically, the researchers

concentrated on timing issues ranging from the purchasing and delivery of inventory to the work
schedules and job assignments of employees. A significant outcome of this work is the
promotion of coordinated activities throughout the purchasing and procurement process such that
schedules can be expanded to include more specific details for improving the resource utilization
across the organization. The models were tested in a hospital in Montreal, and hospital managers
believed that the schedules generated by the models were both efficient and well balanced.
Again, this evaluation furthered this approach as a favorable method to improving inventory
management activities in the healthcare industry.
2.2 Quantitative Modeling in Healthcare SCM and Inventory Control
One of the primary goals of our initiative is to model and improve inventory management
practices being employed by hospitals. There are a number of relevant inventory models in the
current body of supply chain literature that are related to this effort, and those models are
examined and described herein. In this review most of the attention will focus on models
addressing multi-item, single-location and multi-echelon coordination given the unique
conditions germane to the industry considered. Inventory models have been collected that have
been formulated to address healthcare problems, have been applied, or have a high relevance to
healthcare related inventory management.
After summarizing the specifics of inventory control policies and planning frameworks in
hospital pharmacies, the demand specifics are described in this section. A major portion of this
section is devoted to inventory modeling; models of specific interest are multi-item, singlelocation, and single-item, multi-location models. Literature associated with each of these
modeling approaches is discussed, as well as, how these models have been employed in the past.

24

2.2.1 Inventory Control Policies and Planning Framework in Healthcare SCM

Efficiently managing pharmaceutical inventory systems requires another approach than a
continuous review reorder point model. There are at least three limitations for using the
continuous replenishment model in the context of healthcare supply systems: (a) the model does
not account for the limited human resources, (b) it does not account for physical storage
capacities, particularly the one at the CS level which is critical in most hospitals and (c) the
decisions are only based on costs, not accounting for inventory control activities and their
restricted capacity. Each of these limitations is elaborated on below.
Continuous replenishment inventory management is usually not applicable in hospitals
since visits to CUs take a lot of the employees time. To make it more efficient, tours are made
such that several CUs are visited consecutively for rounds of stock control or supply distribution.
The periodic replenishment model seems more appropriate for healthcare organizations as each
CU is replenished according to a schedule instead of when reaching a reorder point. However,
with this inventory management approach, the frequencies of the visits are determined based on
the order period and the economic order quantity (EOQ). Storage capacities at CUs and the CS
are important factors when deciding how frequent a CU should be replenished and how frequent
a supplier should be called. An undersized CS and open shelves able to contain very limited
quantities of the voluminous products is often the reality most hospitals have to face. This
situation calls for more frequent orders to suppliers and/or keeping more stocks at the CU than
these suggested by the EOQ model in order to satisfy requirements and respect capacities. In that
context, CU's stocks are kept higher to compensate for a small CS. Thus, greater volumes of
items can be stored on-site by keeping these extra quantities. Another reason for keeping more
stock in the overall system is that stockouts are time expensive for the supply service that needs
to make extra emergency visits, internally referred as hot-picks, to CUs but also for the medical
25

staff who waste their time making extra calls to the supply department or chasing the products at
other CUs.
One of the distinct features of material management in a hospital is the use of a periodic
review par level (or order-up to level) servicing approach. A major issue in setting par levels for
various items in a healthcare setting is that these levels tend to reflect the desired inventory levels
of the patient caregivers rather than the actual inventory levels needed in a department over a
certain period (Prashant 1991). In most cases these par levels are experience-based and
politically driven, rather than data-driven. This poses a problem for warehouse managers since
the inventory they hold is typically based on aggregate hospital demands while requirements of
departments when aggregated are not in line with such estimates. The literature analyzing the
setting of optimal par levels and review periods for multiple echelons draws upon prior work in
multi-echelon inventory systems which are discussed in the next sections.
On the quantitative side, among one of the first healthcare inventory models was
published by Michelon et al. (1994) who developed a tabu search algorithm for scheduling the
distribution operations in a hospital. In their application, the number of replenishment visits is
given a priori, and the problem consists of finding the schedule that minimizes the number of
carriers given several time windows and practical constraints. Another investigation
concentrating on this type of inventory control situation was Banerjea-Brodeux et al. (1998) who
looked at an application of a routing model to match the different CUs to be visited by a laundry
department in a hospital. Here the authors used a combination of quantitative techniques and
common sense to minimize the number of routes while respecting the volume capacity constraint
of a cart.
Supply chain management can be fit into a planning framework. Vissers et al. (2001)
defined a planning framework for hospitals. One of the primary objectives of this research was to
26

balance service and efficiency throughout the organization. As identified by the investigators,
there are three characteristics that fit hospital production control settings. They are as follows:
(1) a demand that is, generally speaking, larger than supply, (2) by restrictions on supply defined
by contracting organizations, and (3) by higher patient expectations on service quality (Vissers
et al. 2001). As with many organizations the quandary then becomes how to maximize resources
while still achieving higher levels of customer service. The developed framework consists of
five levels of planning and control: patient planning and control, patient group planning and
control, resources planning and control, patient volumes planning and control, and strategic
planning (Vissers et al. 2001). The researchers describe the framework, as well as explain
instances where current hospital policies and practices deviate from their suggestions and where
changes may benefit the organization.
Another attempt is to include SCM into an Enterprise Resource Planning (ERP)
framework. Implementation of ERP in healthcare is very challenging due to the special
characteristics to consider. van Merode et al. (2004) explain the planning function of the hospital
environment and identify facets of the healthcare industry that make ERP implementations and
utilization more difficult. Some attention is given to explaining areas in which ERP can or
cannot be used. According to the authors, hospital functions should be divided into a part that is
concerned only with deterministic processes and a part that is concerned with non-deterministic
processes (van Merode et al. 2004). It is their contention that the deterministic process can be
handled very well by ERP.
2.2.2 Demand Specifics for Hospital Pharmacy
The usage of pharmaceutical products in hospitals has specific characteristics. Typically,
40 to 60% are high demand items that are relatively easy to handle based on usage statistics.
Though, much care is to be taken on the frequent changes in the used medication, on the so
27

called formulary changes. New items come frequently, and an initial qualitative forecast is
required. A large proportion of the items are only rarely used, and the usage times and quantities
are changing. Another group of items is used by patients for only a few days and finished
afterwards. Here the auto-correlation of the demand time series is challenging to consider.
Intermittent demand refers to the usage pattern of items that have extended time gaps
separating irregular or sporadic periods of demand. If the quantities of the usage are also
variable we have the so-called lumpy demand that is a major challenge in pharmacy inventory
management. The study of intermittent and lumpy demand has typically been approached from
either a forecasting or inventory control perspective with a number of studies incorporating both
of these areas to examine practical issues related to inventory system performance. The body of
work discussed in this section provides examples of this.
Sani and Kingsman (1997) provide a good review and compare various periodic
inventory policies like the Normal Approximation (Roberts 1962), Naddors Heuristic (Naddor
1975), Power Approximation (Ehrhardt 1979), and several others. In addition, a variety of
forecasting methods were reviewed with the authors attempting to determine which are best for
low and intermittent demand items. Before discussing specific inventory control approaches or
models, it is important to recognize two fundamental notations commonly used in this literature.
The minimum inventory parameter, or the re-order point, is commonly denoted as s while the
maximum inventory parameter, or the order-up-to value, is represented by S. In situations where
item demand appears to be low or intermittent, variations of the (s, S) form of the periodic
review inventory control system have been promoted as the optimal approach to inventory
management. As the authors state in their review, forecasting demand for such items can also
prove very challenging.

28

The research presented by Sani and Kingsman (1997) examines the management of spare
parts for agricultural machinery, which includes many items with seasonal demand
characteristics displaying vastly different demand from summer to winter months. After
examining the various (s, S) inventory policies with respect to annual inventory costs and
customer service levels, the Ehrhardts (1979) Power Approximation proved to be a good
inventory system for managing items with low overall demand. This method not only scored
well in evaluations of cost minimization but also offered reasonable service levels for items with
low to medium demand. From the comparisons seems that the Croston estimator (Croston 1972)
is one simple forecasting method that performs adequately in most cases (using the exponentially
smoothed inter-demand interval, updated only if demand occurs). One surprising result was that
a basic 12 month moving average of demand yielded adequate forecasting results when
examining both annual costs and service level with exponential smoothing performing the worst
on this measure; however, exponential smoothing performed best on the customer service level
measure alone. It is important to note that both of the aforementioned techniques were utilized
along with updates every review period and the Croston forecast. Finally, the authors concluded
that achieving customer service levels greater that 95% for lumpy demand items is impossible
unless high materials stocks are kept as a means to satisfy this objective.
Hollier et al. (2005) developed a modified (s, S) inventory model to address cost control
issues specific to lumpy demand patterns. This approach integrated a maximum issue quantity
restriction and a critical inventory position as constraints influencing the inventory control
policy. Here the primary objective was to minimize the system replenishment costs. These
authors applied two algorithms, one being a tree search and the other a genetic based, to optimize
the decision variables. The numerical examples and results illustrate the benefits of employing

29

such algorithms, as well as, demonstrate the utility of the maximum issue quantity and critical
inventory position constraints when managing lumpy demand items.
Syntetos and Boylan (2006) employ simulation models to provide an evaluation of
various forecasting techniques, specifically simple moving average, single exponential
smoothing, Crostons estimator, and a new technique introduced in their paper, when handling
lumpy or intermittent demand items. Service level and stock volume were evaluated using a
number of performance measures related to customer service and inventory costs. In this case
the authors reported that the new estimator they developed outperformed other forecasting
techniques as an inventory control method; however, it is important to recognize that the simple
moving average technique yielded favorable results as well compared to the other two methods.
2.2.3 Multi-item, Single-location Models for Pharmacy Inventories
We found few papers that applied multi-item models to control healthcare inventories.
Dellaert and van de Poel (1996) derived a simple inventory rule, a (R, s, c, S) model, for helping
buyers at a university hospital in the Netherlands. The notations of s and S were defined
previously, and there meanings remain the same. Here R represents the length of the review
period (time between orders) while c is the can-order level. Since most items have a joint
supplier and the orders for a certain supplier are always placed on the same day of the week, they
extended an EOQ model to a so-called (R, s, c, S) model, in which the values of the control
parameters s, c and S are determined in a simplistic manner. This approach resulted in
substantial gains, which were observed in improved service levels, reductions in supplier orders,
smaller total inventory levels and holding costs, and substantially lower system costs, for the
participating hospital.
However, we could not find other multi-item inventory applications in healthcare; we
summarize next the models that seem to have the best application potential for this specific area.
30

The majority of results for the multi-item problem are for the deterministic demand case where
the sizes and the multi-product sequence of orders have to be determined. The problem is known
to be NP-complete and therefore many heuristics have been proposed. For an overview on the
approaches, see e.g. Gallego et al. (1996), Hariga and Jackson (1996), Minner and Silver (2005).
The early investigations in stochastic multiproduct inventory problems were generated by
Veinott (1965), Ignall (1966), and Ignall and Veinott (1969). Later in the study of Beyer et al.
(2001) the researchers examined the management of multiple items with a warehousing (storage)
constraint, which was based on the early model of Ignall and Veinott (1969), and generated
results for finite-horizon and infinite-horizon discounted-cost problems. This work demonstrates
optimal policies suitable for the various conditions that occur in healthcare.
Ohno and Ishigaki (2001) examined a continuous review inventory system for multiple
items exhibiting compound Poisson demands and created a new algorithm for determining the
optimal control policy. This alternative method was derived using the policy iteration method
(PIM) and resulted in a substantial decrease in processing time needed to evaluate and improve
the optimal policy. In addition, the new algorithm was tested using three joint ordering policies.
Minner and Silver (2005) analyzed the stochastic demand, continuous review lot-size
coordination problem for Poisson demand and negligible replenishment lead times. A
formulation as a semi-Markov-decision-problem was presented to find the optimal replenishment
policy and several heuristics were suggested and tested in a numerical study. However, the
results only apply for pure Poisson demand and cycle inventories (i.e. without safety stocks).
With the assumption of negligible lead times, safety stocks here serve to avoid the negative
consequences of transaction sizes that exceed available inventory rather than covering against
demand uncertainty over the replenishment lead time.

31

In a recent paper, Minner and Siver (2007) analyze a replenishment decision problem
where each replenishment has an associated setup cost and inventories are subject to holding
costs. The solution of this trade-off results in order batch sizes. The warehouse space for keeping
inventories is limited which generally restricts these batch sizes. A second aspect of their
analysis is that demands are random, here being modeled by a compound-Poisson demand
process. This further complicates the analysis because now, safety stocks and cycle stocks share
the limited warehouse space.
2.2.4 Single-item Multi-location Models - Multi-echelon Coordination in Pharmacy
The literature analyzing the setting of optimal par levels and review periods for multiple
echelons draws upon prior work in multi-echelon inventory systems, which is discussed next.
One of the first investigations in the area of multi-echelon distribution networks was Allen
(1958) who attempted to optimally redistribute stock among several locations. Since then,
numerous works extended this model (see Simpson 1959; Krishnan and Rao 1965; Das 1975;
Hoadley and Heyman 1977). Another significant effort in this area was conducted by Clark and
Scarf (1963) as it was the first study that attempted to generate and depict an optimal inventory
policy in a multi-period, multi-echelon distribution model involving uncertain demand. Other
research in this area has focused on the setting of optimal lot sizes and inventory safety stocks in
a multi-echelon supply chain (Deuermeyer and Schwarz 1981; Eppen and Schrage 1981;
Nahmias and Smith 1994).
Sinha and Matta (1991) and Rogers and Tsubakitani (1991) present modeling studies in
this research domain. Both studies focused on two-echelon inventory systems employing
periodic review under stochastic demand with fixed lead times. In Rogers and Tsubakitani
(1991), the researchers concentrated on finding the optimal par values for the lower echelon such
that the penalty costs were minimized. A budget value acted as a constraint of the maximum
32

inventory investment. The optimal par level is achieved by utilizing a critical ratio, which is
adjusted by the Lagrange multiplier subject to the budget constraint. Sinha and Matta (1991)
focused on minimizing the holding costs of multiple products at both echelon levels with the
presence of penalty costs at the lower echelon. Similarly, the results demonstrated that the
optimal par values for the lower echelon items were obtained using a critical ratio; however, the
holding cost function provided a method for generating the optimal values for the upper echelon.
Nicholson et al. (2004) extend the work of Rogers and Tsubakitani (1991) and Sinha and
Matta (1991) by considering a three-echelon inventory system. Specifically, this research
concentrated on the healthcare sector and sought to compare an internally-managed threeechelon system to an outsourced two-echelon distribution system. Comparisons focused on
inventory costs and service levels. The results suggest that outsourcing the targeted functions
yielded lower inventory costs without sacrificing customer service. As mentioned previously,
the use of periodic review par level servicing at the departmental level (i.e. at the individual care
units) is a unique characteristic to hospital material management; however, it is important to note
that the central pharmacy often operates in a similar manner utilizing another set of par values
reflective of aggregate inventory requirements. Zhu et al. (2005) found similar results in their
study by using numerical simulations and sensitivity analysis of two models. Like Nicholson et
al. (2004), this work demonstrated cost savings with high service levels for the two-echelon
distribution network where item demands were monitored by the distributor and delivered
directly to the departments for use, as opposed to orders being routed through the CS as they
would be in a three-echelon system.
Some models consider the case of lost sales. Nahmias and Smith (1994) examine a twoechelon supply chain where stockouts can penalize the retailer in cases where the customer is
unwilling to wait for their order to be filled. The authors assume instantaneous deliveries from
33

the warehouse to the retailer which is often not the case in practice. Andersson and Melchiors
(2001) executed a similar study that allowed for excess demand to be handled using backorders.
The supply chain in this work was comprised of one warehouse with multiple retail outlets being
serviced. Demand here is independent Poisson processes, and lead times are considered
constant. The heuristic developed by the authors provides a mechanism to evaluate this type of
supply network on elements of cost and service level.
Other research examining the lost sales case is that of Hill et al. (2007), which looked at
inventory control in a single-item, two-echelon system with a continuous review policy. Again,
a central warehouse services several independent retailers and then has its stock replenished by
an external supplier. The system operates such that any excess customer demand at the retailer
can be filled from the warehouse provided the item is in stock; however, any items which are not
in stock result in lost sales. Lead time for the retailer is equal to the transportation time required
to move the item from the warehouse to the specific retail outlet.
The influence of order risk was examined by Seo et al. (2001) in a two-echelon
distribution system. They claim that the service policies should reflect the availability of realtime stock information. Their model adjusts the reorder time on the basis of an approximated
order risk, which is associated with orders that are filled immediately versus those that are
delayed. Results show that this order risk policy performed well when warehouse lead time was
short, where item demand was low, and where there were an intermediate number of retailers in
the supply chain.
In 2003, Axster used an approximation technique to optimize inventory control in a twoechelon distribution network. Items displayed stochastic demand, and the system relied on a
continuous review (R, Q) policy. In this situation holding costs at all locations and backorder
costs at the retailer were assumed to be linear. Axster presents a simplified method for
34

estimating reorder points by using normal approximations for demand at both the retailer and the
warehouse. Lee and Wu (2006) examined a simplified two-echelon supply chain system
comprised of one supplier and one retailer. Here the retailer has a choice between two different
restocking policies: traditional methods (such as an EOQ or periodic review approach) or a
statistical process control (SPC) based method. The results show support for the SPC method in
addressing inventory variation issues and in reducing order backlogs. Benefits are apparent in
the areas of demand management and inventory control, which leads the authors to suggest this
as alternative method for managing supply chain and inventory costs.
2.3 Practical Solutions for Inventory Management of Pharmaceuticals in Hospitals
The following section discusses some practical solutions to issues facing managers when
dealing with prescription drugs. There is a description of some dilemmas hospital personnel face
in filling prescribed medications and the alternative methods employed to satisfy drug orders.
Each of these solutions satisfies various needs of management; however, new concerns also arise
with their use. Finally, a brief overview of Multiple Criteria Decision Making is provided along
with some approaches to decision support.
2.3.1 Current Management Trends and Operations in Pharmacy SCM
Hospitals have typically employed a variety of methods and policies to resolve
pharmaceutical inventory management issues. A few of the more prominent solutions are
discussed in this section to demonstrate the modern approaches seen in industry. Specifically,
outsourcing, VMI, and information system (IS) based solutions are presented.

Outsourcing
As described in the previous literature review, outsourcing has been widely used in the

healthcare industry. Regardless of the specific product, entering into partnerships with suppliers
and distributors for the purposes of combining services have generated benefits for the healthcare
35

providers (see Nicholson et al. 2004; Veral and Rosen 2001; Lunn 2000; Jarrett 1998; Li and
Benton 1996, etc.). The magnitude of resources linked to pharmaceutical inventory and its
management, the desire to shift or reduce pharmacy workloads, and the opportunity to refocus
resources on patient care make this an appealing proposition for healthcare providers. However,
as shown by Rivard-Royer et al. (2002), all parties must benefit from such arrangements to
provide long-term gains.

VMI
Another trend is the growing usage of VMI strategies (Kim 2005). This a specific type of

outsourcing in which pharmaceutical inventories located at various distribution locations (i.e. at

the CUs) around the hospital are monitored by the supplier, in this case the pharmaceutical
company or distributor, and replenished as needed. Currently, several hospitals employ a
continuous review (s, S) inventory control policy. When demand for an item reaches a predetermined minimum level (s), an order is automatically generated and transmitted directly to the
supplier. The supplier, in turn, ships the amount necessary to refill the distribution centers to the
maximum quantity (S). Depending on the specific circumstances, materials can be either sent to
the pharmacy for re-packing and distribution or sent directly to the point-of-service, which
bypasses the pharmacy entirely.

IT-based Inventory Management Solutions

Information systems play a significant role in all of the aforementioned suggested

approaches; however, they are perhaps even more critical in the next solution. Perini and
Vermeulen Jr. (1994) reviewed a number of devices focused on the dispensation of medicines
located around the hospital in the patient care unit (i.e. Lionville CDModule, Meditrol, Argus,
MedStation, Sure-Med, and Selectrac-Rx) with the purpose of replacing the traditional dosage
carts used by pharmacies and to shift control of locally-stored pharmaceutical inventories and
36

controlled substances to caregivers at the point of use. Inventory stored in the CUs offer
caregivers the opportunity to dispense medications quickly to patients; however, restocking these
units can take extra time.
These medication-management machines are designed to offer financial and practical
advantages over traditional operating procedures where inventories are stored in a central
pharmacy and then distributed by dose carts as needed. Another benefit of using these local
devices is the ability to quickly create, store, and access point-of-service patient information,
which can also expedite the documentation requirements associated with drugs. The technology
employed by these solutions enhances operating efficiency and facilitates customer care by
lowering the risk of patients receiving incorrect medications. Regardless of the specific solution,
these systems usually offer pharmaceutical administrators the ability to reduce inventory
carrying and other costs, to improve billing and usage information, and to increase staff
productivity by creating a highly-integrated, data-driven information flow.
2.3.2 Multiple Criteria Decision Making (MCDM)
The International Society on Multiple Criteria Decision Making (ISMCDM) has defined
MCDM as the study of methods and procedures by which concerns about multiple conflicting
criteria can be formally incorporated into the management planning process (ISMCDM 2008).
Here solving a problem is associated with selecting the best alternative from a set of
available and viable solutions. The best solution is the one that most closely satisfies managerial
goals given the decision makers preferences in objectives.
Given the complexity of most business operations, this is an active area of research that
serves as a foundation for decision support system development. In this field there are two types
of problems: 1) multiple criteria evaluation problems and 2) multiple criteria mathematical
programming (MCMP) problems. Multiple criteria evaluation is appropriate when there are a
37

finite number of alternatives that are explicitly known in the beginning of the solution process.
For MCMP the number of alternatives is infinite and not countable, and these alternatives are
usually determined using mathematical models. These alternative solutions are only implicitly
known as well.
There are several methods employed to achieve optimization in MCDM. First, when
there is a single decision maker (DM), a rating method can be used such that each criterion is
rated on a scale of 1 to 10. The results are normalized to obtain weights for each criterion.
Another common procedure is to use one of two ranking methods. The first ranking method is
that of paired comparisons of criteria in which the decision maker is asked for preference
information between various pairs of criteria. For example, decision makers are asked to identify
if criterion A is preferred to criterion B, if B is preferred to A, or if they are indifferent in the
choice. However, there could be a problem with the consistency of preferences. There is a
method known as LINMAP developed by Srinivasan and Shocker (1973) that provides for the
comparison of alternatives to determine the optimal weights. The process is designed to
determine the optimal weights that will minimize the error or inconsistency in the decision
makers preference responses. Other methods involve various scaling methods such as simple
scaling, linear scaling, vector scaling using linear programming norms, linear programming
metrics, and using a combination of ideal solution and linear programming metrics.
Beyond these processes for establishing weights for the various criteria in problem
optimization, there are several approaches for determining the optimal solutions of these
problems. Methods that do not include any knowledge of the decision makers preferences are
global criterion and compromise programming (Yu and Zeleny 1975). Goal programming relies
on pre-specified preferences and attempts to minimize the sum of the weighted deviations from
the goals set for each criterion. Problems here would employ linear goal programming (LGP),
38

partitioning algorithms for goal programs (Arthur and Ravindran 1978, 1980), integer goal
programming (Arthur and Ravindran 1980), nonlinear goal programming (Saber and Ravindran
1993, 1996), and intelligent search methods (Kuriger and Ravindran 2005).

39

CHAPTER 3: RESEARCH ENVIRONMENT AND METHODS

In this chapter, attention is given to the research problems and the methodologies
employed to address these issues. The traditional supply and inventory control operations of
hospitals around the world were presented earlier in this dissertation. Here an overview of the
specific research environment is provided, as well as, some details about the IT-based, inventory
management solution currently utilized by both this local hospital and many others. In addition,
there are a number of quantitative and managerial challenges that are readily observed and
expressed by hospital administrators. The focus of this research is to address these issues and
provide quantitative models of varying complexity and differing data requirements that can be
implemented for improving the operational, tactical, and strategic managerial decision making.
3.1 The Supply Chain and the Major Decisions of the Hospital Pharmacy

The basic structure of the supply network of the case hospital includes
within the hospital
o customers (doctors, nurses, patients),
o local depots (Pyxis MedStations at 86 different locations), and a
o central depot ( Tallyst system at one location);
outside the hospital
o wholesaler (Cardinal GPO, one basic supplier) and
o producers (around six major pharmaceutical companies selected by GPO).
The hospital applies advanced technology in controlling medication throughout the hospital.

The drugs are stored in the local depots (Pyxis MedStations) in 86 different areas of the
hospital. Each area has a different selection of drugs. The Pyxis MedStation registers each
transaction date and quantity of demand (withdrawal) and each delivery (refill) and the actual
inventory level. It is connected to the central depot (Tallyst system) through a computer
network.
As the inventory level decreases to the reorder point (min par level) an automatic order is
triggered by the local depot. The order quantity is determined by the order up to level (max par
40

level). The par levels can be selected and fixed for each drug. Currently, the par levels are based
on fixed day-supplies suggested by the supplier GPO. These values are corrected occasionally by
the local pharmacists using so-called experiences; however, no modeling or optimization is
involved in setting these control values.
The demand for each drug is uncertain. The daily usage data is available for each drug
and each local depot for the period of two years. In the drug supply a high service level is
essential. In case of a shortage at a local depot (Pyxis), an emergency delivery is necessary, and
this emergency refill is very costly.
The central depot (Tallyst) must satisfy the orders of the local depots within a day. One
major goal of the hospital pharmacy is to minimize the number of refills (drugs per location) per
day. If the number of refills per day is very large, it cannot be done in two shifts and to have
overtime or an extra shift is difficult and costly. Shortage in the central depot is rare and in that
case the wholesaler (GPO) refills within a day. The inventory holding cost factor is not a major
concern for the hospital because the wholesaler is financing it. However, inventory holding cost
is included later and in an effort to examine the effect of its consideration.
The physical volume (the total space) of the Pyxis system is limited. It consists of
drawers. Each drawer is subdivided by spacers that can be relocated so different cubicles are
constructed and assigned to each drug. Different drugs cannot be stored in the same cubicle.
The total room for the cubicles is fixed.
The main goals of the management are:
provide a high service level for each drug,
minimize the total number of expected refills (orders) per day for a Pyxis, and
use the limited space of a local depot (Pyxis) in the best way by subdividing it to
separate areas (cubicles) for each drug.
The decision variables are the reorder point (the min par level) and the order up level (max par
level) for each drug. The average daily demand and the standard deviation of the daily demand
41

can be calculated from historical data. Also, we know the storage space requirement for a unit of
each drug and the maximum storage space in a Pyxis.
The operational decision problems occur on the item level: How to set the min and max
par levels for each (250 to 300) drug in each one of the 86 Pyxis MedStations?
- In setting the reorder point (min par level, si) there is a major tradeoff:
increase si since in case of shortage a high emergency refill cost occurs which requires the
provision of a high service level;
decrease si since it needs additional buffer inventory that takes space from cycle stock and
consequently more frequent orders and higher refilling cost arise; refilling cost rather the
inventory holding cost is the primary concern for the hospital.
- In setting the order up level (max par level, Si) the major tradeoff:
increase Si to maximize cycle stock so that less frequent orders decrease the refilling cost;
decrease Si because of the limited space for total inventory in a Pyxis MedStation
There is also a connection between safety and cycle stock (si and Si) for each drug, so it
necessitates the joint consideration of si and Si.
On the other hand, there is a tradeoff among the different drugs in a Pyxis because of
the total space limitation. Thus, the tactical decision problem occurs at the multi-item level:
How to allocate the space (by flexible drawer dividers) among the 250 to 300 drugs in each
Pyxis? What is the best allocation strategy for safety and cycle stock? How to set the si and Si
control parameters according to the allocation strategy? These questions indicate various
tradeoffs to consider among the different drugs in a Pyxis. Further requirements include:

The limited space is subdivided to separate areas (cubicles) for each drug.
The cubicles cannot be shared among drugs.
The max inventory level (Si) must fit into the assigned cubicle.
The major strategic decision problems include: What kind of tradeoffs are among refill

workload, emergency workload, and variety of drugs offered (formulary)? What are their
connections to the three key performance indicators: number of orders, the service level, and the
available space?

42

In the subsequent portions we provide quantitative models, approximations and solutions

starting with the operational decision problems and discuss the applicability and managerial
implications of our models for tactical and strategic decisions in the hospital pharmacy.
3.2 Quantitative Models for the Hospital Inventory Management
This portion of the work presents three quantitative models examined in this
investigation. In addition, a number of iterative processes are utilized to determine the optimal
inventory control parameters given the goals of the individual models and the various
constraints. The notations, models, and the solution processes are identified here.
3.2.1 Model 1: A General Multi-product (s, S) Model with Space Constraint and Its
Approximation
The ultimate goal of the hospital pharmacy operation is to find the optimal values of the decision
variables
- si : the reorder point (the min par level) and
- Si: the reorder level (max par level or order up to level) for each drug i (i = 1 to n)
to minimize the total expected refilling (ordering), inventory holding and shortage cost under
volume constraint of the local depots. The total space is subdivided into separated and dedicated
storage areas for each drug, and they must be large enough to hold the max par level for each
drug. The cost factors for item i are
Ki : cost of a refill (order),
hi = r ci: holding cost for a drug with value of ci , and
pi: shortage cost of an emergency refill that is independent of the size of the shortage.
The optimization problem can be formally expressed in

Model 1:
Min

[Ki Ni(si, Si) + hi Hi(si, Si) + pi Pi(si, Si)]

(1)

s.t.

(vi Si ) M

(2)

43

with the notation for each item

Ni(si, Si) = the expected number of orders per period,
Hi(si, Si) = the expected inventory per period,
Pi(si, Si) = the probability of shortage per period,
vi = volume requirement for a unit,
M = total volume of the space available for the n items.
Scarf (1963) expressed the above expected values, Ni(si, Si) and Hi(si, Si), using renewal
functions. The probability of a shortage in a period has been derived by Schneider et al. (1995)
using the steady state distribution of inventory on hand plus on order published in Iglehart
(1963). Since these expressions use the renewal equation, it is only possible to get the exact
solution for the single-item case with specific demand distributions. A further problem, as it was
shown by Wagner et al. (1965), that the cost function, even for a single item, is not convex.
Thus, following the traditional constrained optimization solution for Model 1 using the
derivatives of the Lagrangian with respect to the parameters si and Si, setting them equal to zero
and solving the equations will not necessarily provide the optimum. However, the Roberts
(1962) approximation based on the first derivatives of the Lagrangian produces nearly optimal
solution for a single item cost function. Schneider and Rinks (1989) suggested an approximate
solution for a constrained multi-item problem similar to Model 1, based on the Roberts (1962)
approximation, and by employing a search algorithm they verify the approximate optimality of
the derived policy. This procedure can also be applied as a benchmark in our case, but it is too
cumbersome for our practical application.
There is a stream of publications handling hospital supply problems with some
similarities to our case and another stream of papers handling similar multi-item constrained
problems with no reference to hospital management. The publications in both streams closest to
this situation are listed next.
44

Michelon et al. (1994) published a tabu search method to optimize the distribution of
supplies in a hospital. Dellaert and van de Poel (1996) derived a simple inventory rule for joint
ordering in a university hospital in the Netherlands. Banerjea-Brodeur et al. (1998) looked at an
application of a routing model to match the different care units to be visited by a laundry
department in a hospital. Vendor managed inventory (VMI) has been applied in healthcare since
the nineties. An analysis and the hospital saving potentials of VMI are demonstrated in Kim
(2005). Meijboom and Obel (2007) investigated supply chain coordination facing a
pharmaceutical company with a multi-location and multi-stage operations structure. They
concentrated on the organizational issues. Lapierre and Ruiz (2007) developed modeling
approaches for improving healthcare inventory management by examining the impact of
scheduling decisions on the coordination of supply activities while recognizing inventory
capacities.
Ordering policies for multi-item inventory systems subject to multiple resource
constraints considering deterministic demand were published in Gder and Zydiak (1999). A
stochastic multi-item constrained model is discussed by Beyer et al. (2001) but only for the case
of base stock policy. Ohno and Ishigaki (2001) examined a continuous review inventory system
for multiple items with compound Poisson demands. Here the joint ordering was targeted but no
space or budget constraints were considered. Minner and Silver (2005) analyzed the stochastic
demand, continuous review lot-size coordination problem without safety stock consideration. In
a recent paper, Minner and Silver (2007) analyze a replenishment decision problem where each
replenishment has an associated setup cost and inventories are subject to holding costs. A
second aspect of their analysis is that demands are random, here being modeled by a compoundPoisson demand process. This further complicates the analysis because safety stocks and cycle
stocks share the limited warehouse space.
45

In summary, the published papers handle several aspects of this research case; however,
they fail to consider jointly the challenge of multi-item and joint constraint with demand
uncertainty or are too complex, time and data intensive for our practical application.
As such, another approach is taken to provide a faster approximate solution of Model 1.
The model is simplified by applying straightforward first order approximation for the objective
function (1) and the constraint (2). Consider
L = lead time (in our practical case it is one period), and use the additional notation for
item i
Di = Si si,
Qi = expected order quantity,
di = expected demand per period,
i = standard deviation of the demand per period,
ui = undershoot quantity, expected inventory position below si when an order is placed. It
can be approximated by the first two moments of the demand per period in the form (see
Schneider 1981)
ui ui(di, i) = (di2 + i2)/2di

(3)

Using this approximation, we can express the expected order quantity as function of di and i:
Qi Qi(di, i) = Di + (di2 + i2)/2di

(4)

Thus we have the following simple approximate expressions as functions of di and i:

Ni(si, Si) di / Qi(di, i)

(5)

Hi(si, Si) si - ui(di, i) - Ldi + Qi(di, i) /2

(6)

The probability of a shortage in a period can be approximated (see in Schneider et al., 1995)

Pi(si, Si) 1 -

( x s ) f ( x | L + 1)dx /Qi
i

(7)

si

where
fi (x|L+1) is the demand density function for a period of length L+1.
The solution of Model 1 using approximations (4) to (7) is more straightforward, and it
could be applied to check different practical scenarios in our case. However, the managerial
problem is that the ordering and shortage cost factors are very difficult to provide. The technical
46

problem is that the large number of items (n=250 to 300 per local depot), the nonlinearity, and
the stochastic demand make the solution challenging and time consuming.
The current control levels suggested by the distributor of the Pyxis MedStations
and by the pharmacy supplier (GPO) is a fixed day-supply, Tmin and Tmax policy, using si = di
*Tmin and Si = di *Tmax , independently from the demand variability or storage space requirement.
This simplistic policy results in frequent shortages and emergency refills for some drugs and also
a large number of regular refills putting an overload and/or overtimes for the pharmacy staff.
Additionally, storage space problems occur frequently. Based on experiences the pharmacists
are frequently modifying the fixed day-supply policy, but there is a need for appropriate decision
support in how to modify the si and Si control parameters. To resolve those high priority
problems, a simplified optimization model was formulated concentrating on the key goals.
The changing composition of the drugs stored (formulary) and the dynamic demand
characteristics call for a fast solution on the daily operational level. The two different simplified
models and simple approximate solutions we provide in the next two sections are also
advantageous to provide efficient tactical decision support in managerial tradeoffs and in
allocation strategies for safety and cycle stock. Further, for strategic decisions, the simplified
models provide easy-to-handle tools. First, in Model 2 consideration is given to both ordering
(refill) and holding cost under service level constraint. In the subsequent Model 3, the
concentration is only on ordering (refill) cost minimization under service level constraint. In
both models the space limit constraint is also included. The simple solutions enable using Excel
spreadsheets that are familiar and easy to interpret by the pharmacists controlling the system.
3.2.2 Model 2: Optimal Allocation Based on Ordering and Holding Costs
In the inventory management literature the difficulty of providing the appropriate cost
factors is well known, especially the shortage cost is hard to quantify. This is valid also in our
47

pharmacy case, so we follow the common practice and consider a service level constraint instead
of shortage cost in the next two models. Since the shortage means an emergency refill in our
case with a fixed cost (depending only on the number of shortage occasions per days and not on
the amount of the shortage) it is appropriate to consider the so-called service level which is the
chance that there is no shortage in a period (day). Using this service level Model 1 can be
modified and simplified into the following form.
Minimize the total refills (orders) cost plus the inventory holding cost for a Pyxis,
subject to the constraints:
-

the service level (chance of no shortage) for each drug is high (at least ),
the total space needed for the maximum possible inventory level for all drugs is not more
than the available total space of the Pyxis MedStation (M).
Using the notation, expressions and approximations (3) to (6) from the previous section,

we can formulate

Model 2:

s.t.

Min K i Qi + hi Hi (si , Si )

(8)

Prob (shortage for drug i) 1-,

(i = 1 to n)

(vi Si ) M

(9)
(10)

The solution of Model 2 is still quite complex because of the several variables, the
nonlinearity and the stochastic constraints. To simplify the solution, we consider the two types of
constraints, service level (9) and space (10) constraints, separately and also the two sets of
decision variables (si and Si) separately and solve the optimization Model 2 iteratively, using
Power Approximation for handling the service level.
An iterative solution is applied based on two embedded iteration procedures for the
optimization of Model 2. The specific Power Approximation formula derived in Schneider
48

(1978) provides a good approximation of the reorder points, si, providing the required service
level .

si = di (Li + 1) + p(yi )i,Li+1

(2i /d i 1) (1.95269 + 6.39059y i )

(1 + 21.17036 y i )

(11)

with (x) = max (x, 0) and p(yi) being a rational function of yi, where yi

yi =

(1)Q i

(12)

2
i,L i +1

depends on the service level, , and also on the Qi values.

To initialize the iteration, we first set the value of the expected order quantity Qi using the
Economic Order Quantity (EOQ) formula, which is noted as Qi'.

Qi =

2d i K i

(13)

hi

Using the above Qi values and the fixed service level we calculate the appropriate si = si(0)
values from Power Approximation (11). The resulting decision variables provide the
approximate optimal solution of Model 2 if the space constraint (10) is fulfilled. Using the
notation of the previous section we get
Si = si + Di = si + Qi - ui

(14)

for the optimal parameters. In the case the approximate optimal par levels, si, Si, (i=1,,n) are
provided by si = si(0), and (14) using the approximation (3) for the expected undershoot
quantities, ui.
If the space constraint (10) is not fulfilled, the above solution is not feasible. That means
the equation
(vi Qi ) M - [vi (si- ui)]

(15)

49

is not fulfilled, and there is a need to decrease the Qi values so that the required storage space for
order quantities (vi Qi ) such that there is less than the remaining total free space for cycle
stock
M = M (si) = M - [vi (si - ui)]

(16)

To find the optimal Qi values that fit into the remaining space, M, defined by (16) we
formulate the sub-problem of minimizing the ordering and cycle stock inventory holding cost
under the remaining total free storage space. In this sub-problem, we fixe the reorder points, si =
si(0) and the resulting M = M0 expressed in (16) using si = si(0).
Sub- problem 2A:
K i

Min

di

Qi

+ hi

vi Qi M

s.t.

Qi
2

(17)
(18)

This sub-problem 2A cannot be solved directly, but the iterative solution procedure
suggested by Ziegler (1982) can be applied.
To initialize the iteration, we use the Economic Order Quantities (EOQ), noted before by
Qi', which also provides the upper bounds, Q(u), on the order quantities for the iteration.
The overall volume of using these Q values is
V = vi Qi

(19)

Here we have the decision rule as described before, if V' M, stop. Otherwise, go to step 1 of
the iterative solution.
In the first iteration step we calculate the lower bound, Q(l), in the form Qi", as

Qi " =

i =

di Ki

Qi

(20)

and calculate values for i, for i = 1, , n, where

v i (Q i ")2

hi

2v i

(21)
50

Once the values of i have been calculated, set ' = min i and " = max i.
In the second iteration step, average the min and max values as determined in step one
to set the overall .
=

(22)

In the third iteration step, we calculate a new Qi using the provided in step two, which
is noted as Qi() for i = 1, , n.
d K

Qi () = h i i
+ v
2

(23)
i

and recalculate V using (19) and substituting the new Qi() for the original Qi. Next, employ the
following decision rules.
If V< M, then set " = and Q(u) = Qi().
If V> M, then set ' = and Q(l) = Qi().
In the fourth iteration step, calculate the functions for the upper and lower bounds as
FQ(u) = K i
FQ(l) = K i

di

(u )
Qi

+ hi

di

(l) + hi

Qi

(u )

Qi

(l)

Qi

(24)

(25)

and use the stopping rule for the iteration

FQ (u ) FQ (l)
FQ (l)

(26)

If the relative difference is determined to be less than or equal to the preset acceptable
error, which was 0.01 in our case, stop. Otherwise, continue the iteration beginning at step two
and repeat this process until convergence. Once the optimal Qi values are obtained, move to the
second stage of the iterative solution and use these Qi values to adjust the reorder points and find
51

the si values providing the required service level, , for each item. With the new si = si(2), we
recalculate M = M2 and solve the sub-problem 2A again with the iteration process above. The
resulting Qi values may be different than those determined in the prior step in which case new si
= si(3) values are used for new iterations until the difference in si and Qi is smaller than the preset
accuracy limit (E = 0.01). The proof of the convergence of the above iterative procedure is in
Ziegler (1982).
3.2.3 Model 3: Optimal Allocation Based on Ordering Cost
In this practical case, the major goal of the hospital pharmacy is to minimize the refill
workload. Inventory holding cost is marginal and is not even considered currently in the Pyxis
database. This is motivation to simplify Model 2 and provide a simple, goal-oriented decision
support tool for the pharmacy management. So, next consider the simplified constrained
optimization problem.
Minimize the total number of expected refills (orders) per day for a Pyxis, subject to the
constraints:
-

the service level (chance of no shortage) for each drug is high (at least ), and
the total space needed for the maximum possible inventory level for all drugs is not more
than the available total space of the Pyxis MedStation (M).

Model 3:
Min (di / Qi)

(27)

s.t.
Prob (shortage for drug i) 1-,
(vi Si ) M

(i = 1 to n)

(28)
(29)

Despite the objective function of Model 3 being simpler when compared to Model 2, we
still have the problem of having the same large number of decision variables, as well as, the
nonlinearity and stochastic constraints as before. As such, a similar iterative solution is applied,
52

but it proves to be faster and easier to interpret than optimization Model 2. Here we also
consider the two constraints, service level (28) and space (29) constraints, separately and also the
two sets of decision variables (si and Si) separately and solve the optimization Model 3
iteratively, using the same Power Approximation as before for handling the service level. This
method is very straightforward, intuitive and sheds light on the simple structure of the optimal
allocation of the safety stock and cycle stock separately.
To initialize the iteration, set the reorder point
si(0) = di *Tmin,

(30)

as it is suggested by the pharmacy supplier. Thus

M0 = M0 (si(0)) = M - [vi (si(0) - ui)]

(31)

is the remaining total free storage space for order quantities.

To find the optimal Qi values we formulate the sub-problem of minimizing the number of
orders per day under the remaining total free storage space for cycle stock with fixed reorder
points, si(0) and the resulting M = M0 expressed in (31):
Min (di / Qi)

(32)

(vi Qi) M

(33)

s.t.

Proposition 1: The optimal space allocation for Qi to minimize the expected number of orders
is proportional with the square root of the demand over volume rate, and the optimal
solution of (32) s.t. (33) is
Qi (0) = M/W * (di /vi)

(34)

W = (vi di)

(35)

with notation

53

Proof: Considering the Kuhn-Tucker conditions for the Lagrange function of Model 2
L(x, ) = (di / xi) + [ (vi xi) - M]

(36)

for the optimal x*i and *

x*i L(x*, *)/ xi = - di / xi + * vi xi = 0 for i = 1,...,n

(37)

so if * is known from (37) we get

x*i = (di / *vi)

(38)

On the other hand, if any of x*i is known

* = di / (vi x*i2)

(39)

Since for the optimal solution the capacity constraint is active

(vi x*i) = M

(40)

Substituting (38) into equation (40) and using the notation

W = (vi di)

(41)

the optimal value of * can be expressed

* = (W / M)2

(42)

and substituting (42) into (38) provides for i = 1,, n the optimal
x*i = (di / *vi) = [M di ] / [W vi] = M/W (di/vi)

(43)

In the first iteration step adjust the reorder points and find the si(1) values that provide
the required service level, , for each item. Here the simple Power Approximation formula
derived in Schneider (1978) expressed in (11) and (12) is employed.
The modified si(1) values will change the available space for cycle stock according to
M1 = M1 (si(1)) = M - [vi (si(1) - ui)]

(44)

which provide the next iteration of

Qi (1) = M1/W * (di /vi)

(45)
54

The iteration for si(2) results from Power Approximation (11) and (12) applying Qi = Qi(1).
Continue the iterations until convergence. According to formula (11), there is a minor effect of
Qi on si, thus the convergence is very rapid; by our experiences two-three iterations are sufficient
to get within 0.1% range.
The approximate optimal control parameters for item i (i = 1 to n) are

min par level, si, is the last iteration expressed in (32) and (33) ;

max par level, Si , is resulting from the last iteration (34) for M
Si = si + Di

(46)

using the approximation with expression (31) for W

Di =

v i
i

d 2i + 2i
2d i

(47)

3.3 Comparison of Optimization Efforts for Model 2 and Model 3

In this section we examine the efforts required to achieve the optimal order quantities, Qi,
for the multi-item case using the sample of 70 items in a Pyxis. First, Model 2 examples are
presented to show the changes in iterations that occur as the optimal order quantities are reached.
Second, an illustration of Model 3 is presented to demonstrate the simplicity of calculations and
the quick nature of optimization using this approach. This is done to contrast these techniques,
as well as, to show the steps and efforts necessary to complete the iterative processes.
3.3.1 Demonstration of Model 2
Model 2 involves a multi-iteration, multi-round process to achieve optimization. The
specifics of the model and this process are presented in Chapter 3. Here we see the precise
results obtained as the iterations progress. Model 2 required several rounds iterations with a
number of iterations within each round. Specifically, 2 rounds of iterations were required with
55

11 iterations necessary in both rounds to achieve convergence. The results below reflect
outcomes of final round of iterations, which ultimately provide the optimal Qi values with the
required accuracy.
To begin the iterative process a comparison must be made between the M value provided
by Model 3 and the overall volume, V, required to house the items in the sample. As shown in
Table 3-1, the space required was determined to be 17874.59 and exceeds our initial M1 value of
4859.79, which prompted the start of the iterative process. With each step of the iteration, the V
is adjusted by the Qi values determined in the previous step in an effort to minimize the
difference between the space needed and the space available, M. The magnitude of these
corrections are very large early on in the process because of the tremendous discrepancy between
V and M1, but this volatility diminishes as optimization is achieved.
Table 3-1: Changes in Overall Volume
V'
V0
V1
V2
V3
V4
V5
V6
V7
V8
V9
V10
V11

Value
17874.59
757.39
1063.28
1484.05
2052.61
2801.62
3753.19
4898.67
4207.27
4511.35
4692.29
4791.93
4844.36

% Change
---2260.01%
28.77%
28.35%
27.70%
26.73%
25.35%
23.38%
-16.43%
6.74%
3.86%
2.08%
1.08%

The values of V shown in Table 3-1 are changed by constantly adjusting the lambda, , values
used in calculating the upper and lower bounds of Qi. These values are available in Table 3-2.

56

Table 3-2: Values of Lambda

'
0.005
0.005
0.005
0.005
0.005
0.005
0.005
0.052
0.052
0.052
0.052
0.052

Initial
Iteration 1
Iteration 2
Iteration 3
Iteration 4
Iteration 5
Iteration 6
Iteration 7
Iteration 8
Iteration 9
Iteration 10
Iteration 11

Values of
"
5.980
2.993
1.499
0.752
0.379
0.192
0.099
0.099
0.075
0.064
0.058
0.055

2.993
1.499
0.752
0.379
0.192
0.099
0.052
0.075
0.064
0.058
0.055
0.054

As explained, the upper and lower bounds of Qi are adjusted with each iterative step.
Table 3-3 summarizes the average percent changes in Qi observed across all items in the sample.
In this case the upper bound of Q was corrected between 24% and 29% in each of the first five
iterative steps. At that point the value of V, as shown in Table 3-1, prompted an adjustment of
the lower bound by 23%. In steps 7 through 11, the corrections shifted back to the upper bound.
At that point, the iterations were halted as a result of the stopping rule.
Table 3-3: Average Percent Change in Q with Iteration

Step 0
Step 1
Step 2
Step 3
Step 4
Step 5
Step 6
Step 7
Step 8
Step 9
Step 10
Step 11

Q' = QU
--29%
28%
27%
26%
24%
0%
10%
6%
3%
2%
1%

57

Q" = QL
--0%
0%
0%
0%
0%
23%
0%
0%
0%
0%
0%

To further illustrate this process an example is provided at the item-level. The drug with
the highest daily demand was selected to demonstrate the convergence to the optimal order
quantity for that item. This is shown in Figure 3-1.

Figure 3-1: Convergence to the Optimal Q for a Sample Drug

The specific results for this drug are provided in Table 3-4. This table supplies both the
upper and lower bounds of Q throughout the iterative process and the observed percentage of
change resulting from each step. As expected, the pattern is similar to that presented in
Table 3-3.
As described earlier in this chapter, the stopping rule for the iterations within each round
is set such that the percent difference between the objective functions of the upper and lower
bounds of Q must be below a preset error. Here the iterations were halted when the error was
less than 1% (i.e. 0.01). Table 3-5 displays the results from the actual iterations in Round 2. As
shown, the convergence is reached by individually adjusting the upper or lower bounds as

58

dictated by the procedure. Gradually, as the iteration evolves, these functions migrate until the
difference is small enough to satisfy our stopping rule.
Table 3-4: Optimization of Q for a Sample Drug

Step 0
Step 1
Step 2
Step 3
Step 4
Step 5
Step 6
Step 7
Step 8
Step 9
Step 10
Step 11

Q' = QU

Q" = QL

QU % Change

QL % Change

16.66
23.53
33.19
46.67
65.29
90.41
90.41
103.00
111.64
116.86
119.77
121.30

144.34
144.34
144.34
144.34
144.34
144.34
122.90
122.89
122.89
122.89
122.89
122.89

--29%
29%
29%
29%
28%
0%
12%
8%
4%
2%
1%

--0%
0%
0%
0%
0%
-17%
0%
0%
0%
0%
0%

Table 3-5: Functions of Q for Model 2 Iterations

F(QU)
Initial
Iteration 1
Iteration 2
Iteration 3
Iteration 4
Iteration 5
Iteration 6
Iteration 7
Iteration 8
Iteration 9
Iteration 10
Iteration 11

F(QL)

2332.92
1684.32
1236.70
932.38
729.55
597.80
597.82
558.56
537.45
526.45
520.82
517.97

775.42
775.42
775.42
775.42
775.42
775.42
515.10
515.12
515.12
515.12
515.12
515.12

Difference
200.86%
117.21%
59.49%
20.24%
-5.92%
-22.91%
16.06%
8.43%
4.34%
2.20%
1.11%
0.55%

This process is performed during each round of the iterative process. After the first
round, the final Qi (Qi = QL) are used in the Power Approximation formula (11) to calculate the
corresponding reorder points, si, for each item in the sample. These new si values allow for a
59

new M1 value to be established, which initializes the next round of iterations. The values of M1
used to initiate the model are presented in Table 3-6.
Table 3-6: Values of M for Iterative Rounds
Round 1
4859.79

M1

Round 2
4860.14

% Difference
0.01%

The rounds of iterations continue until the difference between Qi and si values is small
enough to meet our stopping rule of less than 1% difference. As mentioned previously, this
specific example needed 2 rounds of iterations with each round requiring 11 iterations to reach
convergence. Upon completion of Round 2, the resulting Qi values are used once again in the
Power Approximation formula to set to determine the corresponding si values for the sample
items. Table 3-7 shows the calculated values of Qi and si for the sample and supports the
decision to stop the iterations after 2 rounds.
Table 3-7: Order Quantities and Reorder Points for Sample
Iteration 1
Drug 1
Drug 2
Drug 3
Drug 4
Drug 5
Drug 6
Drug 65
Drug 66
Drug 67
Drug 68
Drug 69
Drug 70

Iteration 2

Qi(1)

si(1)

Qi(2)

si(2)

122.89
122.62
120.03
51.11
71.48
66.34
24.95
28.75
35.88
18.92
18.77
30.68

59.95
34.23
26.00
25.99
17.86
15.82
2.96
2.19
6.87
5.40
4.48
1.72

122.89
122.62
120.03
51.11
71.48
66.34
24.95
28.75
35.88
18.92
18.77
30.68

59.90
34.21
25.98
25.97
17.85
15.81
2.96
2.18
6.85
5.40
4.47
1.72

60

% Change
Q
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%
0.00%

s
0.08%
0.07%
0.10%
0.07%
0.08%
0.11%
0.09%
0.17%
0.27%
0.10%
0.10%
0.13%
0.11% Average

3.3.2 Demonstration of Model 3

In contrast to the complexity of Model 2, a significant benefit of Model 3 is that it
provides a much simpler and less cumbersome approach to optimizing the order quantities. A
benefit of this approach is the need for relatively few inputs, which satisfies a major obstacle
facing the hospital pharmacy. At present, the hospital and GPO were unable to provide any cost
related data. It simply did not exist within the pharmacy database. In addition, something as
basic as extracting accurate usage data for transactions at the Pyxis units is a very arduous task.
According to the pharmacy staff, usage data can only be extracted for a period of 6 days at a
time. Pulling enough data for accurate analytical purposes is labor-intensive and taxes an
already over-worked staff.
Once the model inputs are identified, the calculations are uncomplicated with a single,
straightforward stopping rule. The model inputs include the average daily demand, the standard
deviation of daily demand, and the unit size along with the current inventory control policy used
by the hospital (shown in Tables 3-8 and 3-9). This information is used to initialize the iterative
procedure.
Table 3-8: Fixed-days Supply Policy of Hospital
Tmin

Tmax

10

The hospital control settings (s, S) provide the initial values necessary to determine the
space available for cycle stock, M0, given the reorder points. From this, the starting values for Qi
are established. These order quantities are then used in the first iterative step to modify the si
values for the sample items, which creates a new value of M, M1. A comparison is then made
between M0 and M1. According to our stopping rule, the iterations continue until the change in
M values from one step to the next is less than 1%. The changes in M values are shown in Table
61

3-10. Using this approach, convergence is usually reached quickly and only required 2 iterations
in this optimization example.
Table 3-9: Inputs for Model 3
Drug

Average

Drug 1
Drug 2
Drug 3
Drug 4
Drug 5
Drug 6
Drug 65
Drug 66
Drug 67
Drug 68
Drug 69
Drug 70

16.63
10.54
8.29
7.60
5.28
4.76
0.70
0.69
0.68
0.68
0.67
0.66

Sigma

Size (vi)

13.10
8.50
6.97
5.54
4.81
4.30
1.24
1.03
2.33
1.67
1.49
0.85

2
1
1
5
1.5
2
1
1.5
1
2
2
1

Table 3-10: Percent Change in Space Available to Cycle Stock (M)

Initial
M

5357.58

% Change

---

Iteration 1

Iteration 2

4904.59

4859.79

9.24%

0.92%

Once the final values of si and Qi are found, the difference between the reorder point and
the order up to level, Di, can be set according to the formula (47) and used to calculate the Si
values for the items, where Si = si + Di. The values of si and Qi from the iterations are shown in
Table 3-11 along with the final par values as determined by this approach.

62

Table 3-11: Optimal Values from Model 3

Initial

Iteration 1

Iteration 2

Drug

si(0)

Qi(0)

si(1)

Qi(1)

si (2)

Qi(2)

Drug 1
Drug 2
Drug 3
Drug 4
Drug 5
Drug 6
Drug 65
Drug 66
Drug 67
Drug 68
Drug 69
Drug 70

49.90
31.62
24.87
22.81
15.84
14.29
2.11
2.08
2.05
2.05
2.02
1.99

117.08
131.80
116.90
50.06
76.17
62.65
34.04
27.63
33.57
23.74
23.55
33.04

60.59
33.53
26.18
26.09
17.50
16.08
2.48
2.23
7.20
4.76
3.96
1.66

107.19
120.66
107.02
45.83
69.73
57.35
31.16
25.29
30.74
21.73
21.56
30.24

61.86
34.36
26.87
26.58
17.98
16.49
2.62
2.32
7.66
5.01
4.16
1.73

106.21
119.55
106.04
45.41
69.10
56.83
30.88
25.06
30.45
21.53
21.36
29.97

Optimal Values
Di
92.73
110.86
98.97
39.59
64.27
52.51
29.42
23.96
26.13
19.16
19.38
29.09

si
61.86
34.36
26.87
26.58
17.98
16.49
2.62
2.32
7.66
5.01
4.16
1.73

Si
154.59
145.22
125.84
66.17
82.25
69.00
32.04
26.28
33.79
24.17
23.54
30.83

3.4 Accuracy of the Approximations

We examine the accuracy of our approximations for the three key performance
indicators:

the expected total number of orders (daily refills),

the maximum space requirement for the total cycle stock, and
the average service level.

We compare our approximations based on Model 3 with the result of using simulations of daily
transactions. We applied various parameter settings and in the simulation we used the real
demand observed at a particular Pyxis MedStation of the case hospital for a year.
For illustration, in Table 3-12, we summarize our results based on the same case example
we used in the previous part of this Section for the selected 70 drugs in the particular Pyxis
with the highest demand rate. We examine the performance characteristics calculated and
simulated for the three inventory control policies: the supplier suggested (GPO), the one
modified by the hospital (HM), and our approximate optimization Model 3 (OM). The maximal

63

space available (M) is set to be the same for each policy, which is equal to the volume that is
currently used by the HM policy for the total cycle stock.
Table 3-12: Accuracy of the Approximations - Comparison of the Calculated and
Simulated Key Performance Indicators
Simulation Results
Inventory Policy
GPO
HM
OM

Refills
6.20
5.86
4.52

Max Space
5845.75
5898.50
5691.50

SL
99.95%
99.98%
99.98%

Model Predictions using our Approximation

Inventory Policy
Refills
Max Space
GPO
6.03
5860.75
HM
5.74
5931.50
OM
4.47
5921.50

SL
99.89%
99.84%
99.94%

Percent Difference
Inventory Policy
GPO
HM
OM

SL
-0.05%
-0.14%
-0.05%

Refills
-2.86%
-2.05%
-1.12%

Max Space
0.26%
0.56%
3.88%

As the above illustration in Table 3-12 shows the percent errors are below the 4% margin
with significantly lower errors on the measure of average service level. Overall, these results
effectively demonstrate the accuracy of our approximations. However, it is important to note
that we observed that below 99% there is a tendency of underestimating the average service level
for low demand rate. We discuss the reason for this phenomenon in the next paragraph. The daily
usage rate is in the 16.63 to 0.66 range for the 70 items we considered. 35% of the items are
below the 1 unit daily usage rate. As our detailed investigations revealed, these low-usage items
are the main sources of the approximation errors. Disregarding them, the approximation error
decreases below 1.2% that is well acceptable for our practical situation.

64

Figure 3-2: Approximation of Target Service Level

Figure 3-2 shows the average service level simulated for the 70 items for different target
service levels. While the accuracy of the average service level approximation we achieved using
simulation is good for the 99% target service level and above, there is a large error at the lower
average service levels. To explain this phenomenon we need to examine service at the item
level. This observed error can be explained by the absence of shortages for many of the low
usage rate items. 76% of the low-usage items showed no shortage during a year period. The
remaining 24% of these items typically had one shortage occasion and displayed high demand
variability with all shortage items having a standard deviation of daily demand that was 2 to 3
times the amount of average daily demand for the item (this is the so-called lumpy demand).
Correcting the error at the lower average service levels would allow for an increase in stockouts
and more accurate service level estimate. However, this is impractical in our case given the
expense of emergency refills and sensitivity of patient care. On one hand we can conclude that
low volume items and lumpy demand items need better service level approximation, but on the
other hand this is not so important for our practical case because their total safety stock
contribution is small.
65

CHAPTER 4: ANALYSIS AND MANAGERIAL INTERPRETATION

In this chapter the focus is on describing the decision support tool developed for this
project and the managerial interpretation of the analysis performed. To facilitate this research
and to make this endeavor more applicable and beneficial for the stakeholders at the participating
hospital, a great deal of consideration was given to their current operations and the manner in
which inventory data is utilized. First, the rationale for the decision support tool and software
selection is given. Second, comparisons are made between the efforts to achieve optimization
using Models 2 and 3. Third, a comparison is made to evaluate the estimated inventory costs of
employing the current hospital policy, Model 2 results, and the Model 3 control values. Finally,
the operational, tactical, and strategic implications of the findings are explored and used to
illustrate the practical value of this work. To that end, a number of managerial tradeoffs are
considered along with comparisons amongst allocation strategies.
4.1 Application for Managerial Decision Support
For this project a modeling application and simulation were developed using MS Excel.
There are a number of reasons this software was chosen. First, the demand and usage data is
available for extraction and manipulation in a file format compatible with this program. The ITbased solution currently employed by the hospital to monitor and control pharmaceutical
inventory works quite well with this software. Second, pharmacy administrators are very
familiar with this application, as well as, with the MS Office Suite. MS Excel is used
extensively by the staff and is one of the most commonly used spreadsheet applications around
the world. As such, this was a logical application to utilize for the program and subsequent
analysis.

66

Data from the identified Pyxis MedStation is extracted at the transaction level, which is
then manipulated into daily usage values. Once the data is in this form, the average daily
demand and the standard deviation of daily demand is calculated for input into the quantitative
models. The formulae and iterative processes (as described in Chapter 3) are incorporated into
the application such that changes in control parameters are readily available for review
throughout the procedure. An added benefit of having the daily usage data for every item in a
local depot is the ability to verify the accuracy and utility of the calculated par values. In
addition, now that the application has been tested and the accuracy established, the iterations of
the models can be automated in MS Excel by setting a predetermined stopping value (i.e. 0.01)
within the program when the difference between iterations is small. This is ideal for managers
that have limited amounts of time to devote to optimizing operating settings and that have little
interest in the steps necessary to achieve the final values.
Ideally, this application would be incorporated into the workings of the pharmacy, which
is a relatively easy task, due to the utility of the program and the ease of its use. Furthermore, to
achieve a greater potential impact on the critical performance indicators, mechanisms for
updating inventory control parameters are easily added. For example, forecasting techniques,
such as exponential smoothing or moving averages, might prove useful in monitoring product
demand and adjusting the formulary to take advantage of the tradeoffs explained here. Also, the
continuous review of inventory at any given Pyxis MedStation and across all of these machines
allows for real-time evaluation of the control values and facilitates swift changes and
adjustments to settings as necessary.
4.2 Levels of Decision Support
This section identifies the manner in which the simplified approach of Models 2 and 3
supports decision making at multiple levels. First, the operational decisions are discussed.
67

Second, the tactical considerations are presented along with the utility of this decision support
tool in analyzing managerial tradeoffs. Third, a breakdown of the strategic implications is
provided.
4.2.1 Operational Decision Support
At this level of decision making, the focus is on the management of individual items.
The high service level requirement dictates a high reorder point, si, be used to maintain
healthcare standards and to avoid expensive emergency refills. Additionally, the order up level,
Si, for each item must be reduced to accommodate the other products in the Pyxis given the
space constraints of the storage unit. The reduced space for cycle stock generates a need for
additional daily refills, which results in higher refilling costs and workloads for the pharmacy.
Any changes at the item level can impact the operations and workloads associated with the local
depot. Managers needed help in evaluating their current practices and in improving these
procedures.
The decision support tool developed for this research project serves as a model for
managers. It allows them to examine changes in the formulary or item usage, to evaluate options
for modifying the control parameters, and to choose the par values that best fit hospital and
management criteria. In addition, it allows for quick, simple analysis of the managerial tradeoffs
associated with the pharmacy and storage unit capacities. Greater details on this issue are
provided later in this chapter.

Cost Comparison of Allocation Strategies

As described above, two methods for determining the optimal par levels and order

quantities in multi-item, single-location settings under constraints have been explored. Here
comparisons are made between the current hospital inventory policy (HM) and the alternative
approaches of Model 2 and Model 3. This is followed by comparisons between the two
68

alternative approaches. As a reminder, the hospital currently uses a modification of the fixed
day-supply policy suggested by the GPO. Model 2 is designed to find the optimal allocation of
space across all items being considered by minimizing the sum of holding and ordering costs.
While Model 2 focused on the issue of holding and refill costs, Model 3 was established to
satisfy an exact desire of the hospital pharmacy to minimize the daily refill load without
consideration of holding costs. Specifically, Model 3 minimizes the total number of expected
refills (orders) per day for a Pyxis, subject to the service level and storage capacity constraints.
Although accurate cost data was unavailable from the hospital, unit prices were obtained
using a well-known, online pharmaceutical vendor. Despite not having exact holding cost values
for the sample drugs, surrogate values were used for comparative purposes and to better
understand the cost implications of these policies. Figure 4-1 illustrates the costs as calculated
using actual product prices. Here the cost implications of such policies are easily observed.

Figure 4-1: Cost Comparison for Allocation Methods

69

As shown in Figure 4-1, the holding cost, refill cost, and total cost of refill and holding
are substantially different for the three policies. Holding costs are nearly identical for HM and
Model 2 with the Model 3 policy resulting in holding costs around 40% higher than the other
strategies. Refill costs are substantially higher for the HM policy than for either Model 2 or
Model 3 with the refilling cost being 71.1% and 83.9% higher respectively. These higher refill
costs result in a much higher total cost for the HM policy and is evidenced by a 52.8% higher
cost than Model 2 and 36.4% higher cost than Model 3. These differences are summarized in
Table 4-1.
Table 4-1: Percent Differences in Model Costs to Hospital Policy

Model 2
Model 3

Percent Difference
Holding Cost Refill Cost Total Cost
1.3%
-71.1%
-52.8%
41.4%
-83.9%
-36.4%

In continuing with these comparisons, it is useful to examine the differences between the
two approaches suggested as alternatives to the HM policy. Again, the two models are evaluated
on the basis of holding cost, refill cost, and total cost of holding and refill. The results are
available in Table 4-2.
Table 4-2: Cost Comparison of Model 2 and Model 3

Model 3

Percent Difference
Holding Cost Refill Cost Total Cost
40.6%
-7.5%
10.7%

Model 2 provides 10.7% lower total cost of refill and holding with much of that due to
the 40.6% lower holding cost realized by the model. Again, this is understandable given the
cost-focus of this approach. The inclusion of prices allows for allocation to be based on item
70

cost and ensures that higher costing items are kept in smaller quantities, which enables the
savings identified previously. On the other hand, Model 3 requires a 7.5% lower refill cost,
which supports the desires of hospital administrators.
It is important to reiterate that neither the hospital nor the GPO could produce accurate
cost data to support this research. As such, the cost analyses are performed primarily to serve
three purposes. First, when holding and ordering costs are unknown, Model 3 provides a
simplified technique for approximating the optimal order quantities, Q, that minimizes the
number of refills. It is likely that many organizations operate without accurate cost information
similar to our case hospital, so Model 3 offers an opportunity to assess policies and possibly
improve upon them. Second, as with this pharmaceutical case, holding cost was not a primary
concern for the pharmacy. Model 3 addresses their need to reduce refill activities, thereby
decreasing workloads, while continuing to offer high service levels for caregivers administering
patient treatments. This is not to say that costs are unimportant to pharmacy management, but it
is not the main objective. Third, this analysis clearly demonstrates the significance of having
accurate cost data to improve the cost performance of the supply chain and decision making.
The comparisons with Model 2 reveal the relationship of the holding costs, hi, to setting the
optimal order quantities, Qi, for all items. When holding costs are high, Qi values are set lower
to avoid elevated carrying costs. Conversely, items with lower holding costs can be held in
higher quantities with small penalty. In cases where good cost data is available, the expected
refill size Qi is proportional to the rate of ordering cost, Ki, over holding cost, hi.

Additional Considerations for Improved Operations

It may be unrealistic for practitioners to re-run the approximation methods every time a

change is proposed in the formulary. The demand for drugs changes on a daily basis, and items
are added and removed from the formulary frequently. It may be too demanding to run the
71

iteration procedure each time a change occurs. Thus, some suggestions are offered as support for
improving the day-to-day operations of the pharmacy. After an initial run of Model 3 to identify
the reorder point and set the optimal order quantities, the pharmacy can employ an abbreviated
technique since some of the variables in this equation will remain relatively constant during a
short period of time. The optimal si values are provided by the Power Approximation (11) and
set to satisfy the service level requirement. Using the final values from the initial run for the
space available for cycle stock (see Table 3-10 for values of M), and our calculated W of 131.97
as determined by formula (41), we can provide a quick approximation of Qi for any item
according to formula (45). As explained previously, the optimal space allocation for Qi to
minimize the expected number of orders is proportional with the square root of the demand over
volume rate. Knowing the appropriate si and Qi values, the order up to level can be set according
to Si = si + Qi ui. It is important to note that this is not the exact optimal solution; however, it
will provide a good estimate of the solution until the next optimization run for the Pyxis is
made by the pharmacy.
Similarly, we propose a fast, simple approximation technique using the initial optimized
values of Model 2. Using the final value of (see Table 3-2) from the initial optimization run
and the formula (23) for Qi, a good estimate of the optimal Qi value can be generated. Since the
reordering costs remain fairly constant over short periods of time and an appropriate holding cost
is known, this formula can be applied by simply inputting the proper values of average daily
demand and the unit size requirement. Again, this does not provide the exact optimal solution,
but the simplified nature of this approach makes it a good alternative until another optimization
run is made for all items in the Pyxis. It allows pharmacy workers to make quick
approximations of the optimal order quantity for an item any time a new drug is added to the
formulary or when there is significant change in the average usage of an item.
72

Another important relationship to recognize is the influence that the rate of variance in
daily demand to average daily demand (2/d) has on inventory policy. The current hospital
policy of a modified, fixed-days demand approach means that the reorder points for all items in
the Pyxis are set on average to be 3 days of expected demand. Understanding the influence of
variance in daily demand and incorporating the 2/d ratio into the inventory policy can improve
the operations. When this rate is large with the variance of daily demand being 2-3 times or
more than that of the average daily demand, the pharmacy may consider modifying the policy
according to the Power Approximation (11) and set a higher reorder point and lower the order up
to level (i.e. use a 4, 6 policy not 3, 10). This higher reorder point allows the pharmacy to avoid
stockouts and emergency refills and offers a mechanism for detecting atypical usage patterns. In
addition, the lower order up to level enables the pharmacy to avoid keeping unnecessarily large
inventory of low usage items. On the other hand, when the rate is lower with the variance of
daily demand being equal to or much smaller than the average daily demand, then the hospital
may consider lowering the reorder point and set the a higher order up to level (i.e. use a 2, 12
policy not 3, 10). Since orders are filled within 1 day by both the pharmacy and the drug
supplier, replenishing inventory should not be a problem. Thus, the additional space allocated to
an increased cycle stock limits the refill occasions for these items.
Both of the simple rules above offer improvements to current hospital pharmaceutical
management practices. In addition, according to pharmacy information, technicians have limited
access to usage data and only have the ability to extract 1 weeks worth of usage data at a time.
Since data availability and worker capacity is limited, employing this abbreviated technique and
understanding the influence of demand variability on inventory policy can greatly enhance
pharmacy practices at the operational level. Models 2 and 3 demonstrates the ability to address
both the managerial goals of the pharmacy and to offer significant quantitative support for
73

workers charged with daily tasks in the pharmacy. The pharmacy can make optimization runs
less frequently (i.e. every month or quarter) or only when substantial changes in the formulary or
item demands have occurred.
4.2.2 Tactical Decision Support
In this section discussion shifts to What If analysis and the managerial implications of
the findings demonstrated previously. The benefits and ability of Model 3 to outperform both
GPO and current hospital practices has already been established in the previous chapter. Now,
we demonstrate what happens if we make various changes to the current operating procedures
and conditions.

Safety Stock and Cycle Stock Allocation Strategies

Another important practical significance of Model 3 is to provide simple approximation

to the optimal safety stock and cycle stock allocation strategy. There is also a relationship
between the allocation strategy and operational control: how to select the near-optimal min and
max par levels (s and S). This is shown below.
The safety stock is determined by the choice of the reorder point, s. The reorder point, s,
is the sum of expected demand during the lead time plus the safety stock. The lead time is the
refill time, which is one day in our case. According to the Power Approximation (11), the
appropriate reorder point selection, providing the required service level, depends also on the Q
value. However, this influence is marginal compared to the effect of the and demand
parameters (mean and variance). Based on this observation, we can separate the allocation
strategy for the min par levels from the allocation of the max par levels and provide a simple
approximate strategy for the allocation of s as a function of the demand variability.

74

Considering the demand variability provides a significant improvement in the consistent

allocation of safety stock compared to the simplistic fixed day-supply strategy suggested by the
supplier (GPO) and also to the modified allocations based on local pharmacy experiences. We
summarize the results of the comparisons next starting with the service level comparison.
Table 4-3: Policy Comparison with Constant Space Utilization (M)
Inventory Control Policy

Avg. Service Level

Service Level Range

Avg. daily Refills

GPO Suggested (GPO)

99%

7.15%

5.56

Hospital Modified (HM)

99%

5.14%

5.74

Our Model 3 (OM)

99%

0.44%

4.70

In Table 4-3, we examine the performance of the three inventory control policies (GPO
Suggested, Hospital Modified, and Our Model 3) when the space availability (M) is set to be
the same for all approaches, equal to the value that is currently used by the hospital (HM) policy.
First, we evaluate the three approaches on the issue of service level. As shown above, all three
policies perform very well on the average service level criteria with Our Model 3 (OM)
outperforming the others in the consistency of the service level, measured by the service level
range (max min service level) achieved across all items in the Pyxis unit. With respect to the
number of refills required per day, the OM method outperforms both of the other approaches,
which is a primary goal of our project.
Table 4-4: Policy Comparison with Constant Service Level (SL)
Inventory Control Policy

Space Used

Service Level Range

Avg. daily Refills

GPO Suggested (GPO)

99%

3.94%

6.03

Hospital Modified (HM)

100%

5.14%

5.74

Our Model 3 (OM)

82%

0.44%

5.68

75

Next, we compare the policies when a fixed average service level of 99% is set. Table 4-4
above presents the findings. In this case we notice that while the GPO and HM policies require
essentially the same amount of storage space to achieve the required 99% service level the OM
policy needs only 82% of that space to meet the same requirement. In addition, the OM policy
offers a much more consistent level of service across the items found in the Pyxis. The range in
service level is less than half of a percent (0.44%) compared to 3.94% and 5.14% for the GPO
and HM policies respectively. Further, employing the OM suggested par values resulted in
slightly lower numbers of refills than the GPO or HM policy, which is noteworthy given the
substantial difference in the amount of space used for inventory in this approach, under this
service level constraint. The additional space available for use with the OM policy allows for
greater flexibility in formulary changes and pharmaceutical management. As such we can
increase the space allocation for cycle stocks to use the full capacity of the Pyxis. In doing so,
we exceed the service level requirement and can further reduce the expected number of daily
refills in the Pyxis. These results are summarized in Table 4-5 below.
Table 4-5: Policy Comparison with Constant Service Level and Increased Space
Inventory Control Policy

Avg. Service Level

Service Level Range

Avg. daily Refills

GPO Suggested (GPO)

99%

3.94%

5.56

Hospital Modified (HM)

99%

5.14%

5.74

Our Model 3 (OM)

99%

0.44%

4.70

Finally, we examine the changes in performance indicators if we set all of the average
number of daily refills for each competing policy, equal to the refill number achieved by the
current HM policy. We see in Table 4-6 that the OM policy uses less space (82%) than the GPO
(99%) or HM (100%) inventory policies. All three policies perform well on the measure of
average service level with all methods exceeding the 92% mark; however, the HM and OM
76

policies demonstrate the ability to deliver exceptional service by surpassing the 99% level. As
shown previously, the OM policy continues to offer a more consistent service level at the item
level than either of the other two management policies.
Table 4-6: Policy Comparison with Constant Number of Daily Refills (N)
Space Used

Avg. Service Level

Service Level Range

GPO Suggested (GPO)

99%

92.00%

27.60%

Hospital Modified (HM)

100%

99.84%

5.14%

Our Model 3 (OM)

82%

99.93%

0.44%

Inventory Control Policy

One of the main results from the practical point of view is that the optimal allocation
strategy of the space for the order quantities (cycle stock) of the items is proportional with the
square root of the demand over space rate. Comparing with different allocation schemes (like
proportionally with demand, or volume) it is interesting to verify the value of the right space
allocation in reducing the total number of expected orders per day. We have compared five
different allocation strategies for several examples. Table 4-7 summarizes the average percent
increases in the total number of refills applying an allocation rule that is different from the
optimal one.
Table 4-7: Comparison of the Different Space Allocation
The expected refill size Qi is proportional to

Percent increase in total # of orders per day

SQRT(di/vi) the optimal

0%

Demand rate, di

10%

Unit space requirement, vi

400%

di * vi

177%

di/vi

38%

77

Changes in Holding Costs

In this section we examine the influence the holding cost has in determining the optimal

order quantity for Pyxis items when Model 2 is used as the allocation method. The allocation
depends on the cost rate of Ki / hi where hi = r ci and ci is the unit price for an item. Specifically,
the inventory holding cost rate, r, used to set the holding costs for the formulary items is
manipulated. We select different r values to observe the resulting changes in both the iterative
procedure and the optimal order quantities generated. In the demonstration of Model 2 provided
in Chapter 3, a daily holding cost rate of 0.002 (r = 0.002) was used. In this analysis we modify
this rate to reflect increases or decreases in holding costs and discuss the outcomes.
In Table 4-8, a summary of the changes in the available space for cycle stock, V, is
presented. Here we show the starting values of V as calculated using formula (19) and the
resulting final value of V achieved as a result of the second round of iterations. First, one can
see that smaller rates, which result in smaller holding cost, have a greater starting value of V.
For example, r1 = 0.001 starts with a V = 25278.49 and is significantly larger than V of either r2
(r2 = 0.002) or r3 (r3 = 0.003), which are 17874.59 and 14594.55 respectively. However, the final
values of V reached at the conclusion of the iterative procedure are nearly identical for all three
rates. This provides evidence of the convergence to the value of the available space for cycle
stock and the optimal solution achieved by this allocation strategy. Another notable observation
is that the use of r3 yielded convergence in fewer steps (10) than either of r1 or r2, which required
one additional iterative step to yield optimal values.
Table 4-8: Comparison of Space Available for Cycle Stock (V) at Varying Holding Costs

r1 = 0.001
r2 = 0.002
r3 = 0.003

V'

Vfinal

Iterations

25278.49
17874.59
14594.55

4817.43
4844.36
4867.37

11
11
10

78

Next, we summarize the percent change that occurs from the initial value of V to the
value of V provided by the final iteration. These are available in Table 4-8.
Table 4-9: Percent Change of Space Available for Cycle Stock (V) at Varying Holding
Costs
Vfinal
4817.43
4844.36
4867.37

V'
25278.49
17874.59
14594.55

r1 = 0.001
r2 = 0.002
r3 = 0.003

% Change
80.94%
72.90%
66.65%

As demonstrated by Table 4-9, the holding cost rate influences the amount of change
needed to reach the final value of space available for cycle stock. With an extreme value of r1,
the difference between the initial and final values of V was nearly 81%. As the r value is
increased, it appears that the initial amount of space needed is a better starting value for the
iterative procedure and requires less correction to reach the optimal value. This is shown by the
smaller percentages of change for r2 (72.90%) and r3 (66.65%).
The demonstration provided in Chapter 3 used a rate of r2 = 0.002 to determine the
holding costs of drugs. Here we examine the impact of increasing or decreasing this rate has on
the values of V. In Table 4-9 the comparison is made between r2 and the 2 other holding cost
rates of r1 = 0.001 and r3 = 0.003.
Table 4-10: Impact of Changing Holding Costs on Available Space (V)
Percent Difference (%)
V'
r1 = 0.001
r3 = 0.003

29.29%
-22.47%

Vfinal
0.56%
0.47%

As evidenced by Table 4-10, changes in the holding cost rate and holding cost will
influence the initial estimate of space needed for cycle stock. If the value of r is decreased from
0.002 to 0.001, these results indicate a difference of 29.29% between the starting values of V
79

with the value associated with r1 being that much larger than the corresponding value of r2. In
addition, if the holding cost rate is increased from 0.002 to 0.003, the initial value of V for r3 is
22.47% smaller than that of r2. Again, this supports the assertion that the higher holding cost
rates provide starting values of V that are closer to the actual space requirement of cycle stock.
The above results also support the ability of this allocation technique to converge on the optimal
solution regardless of the holding cost rate and subsequent holding costs. The final values of V
are almost identical for all three rates, which is evidenced by the small differences of less than
0.56% across the three values.
Next, we demonstrate the changes in the optimal order quantity resulting from the
varying inventory holding cost rates. As explained previously, the hospital pharmacy and the
GPO were both unable to provide unit prices or accurate holding cost data. Unit prices were
obtained from a well-known, online pharmaceutical vendor. Table 4-11 presents an example of
the prices used in this analysis.
Table 4-11: Optimal Order Quantities (Qi) at Differing Holding Costs
Q Values with Different r
Drug
Drug 1
Drug 2
Drug 3
Drug 4
Drug 5
Drug 6
Drug 65
Drug 66
Drug 67
Drug 68
Drug 69
Drug 70

Price ci
$
$
$
$
$
$
$
$
$
$
$
$

5.90
17.97
5.43
30.38
25.17
3.96
60.76
5.90
1.00
86.81
86.81
18.23

r = 0.001
116.04
123.35
114.72
49.00
71.59
62.34
27.84
27.28
33.54
20.42
20.26
30.89

80

r = 0.002
122.89
122.62
120.03
51.11
71.48
66.34
24.95
28.75
35.88
18.92
18.77
30.68

r = 0.003
126.75
119.10
122.25
51.89
69.67
68.80
22.53
29.49
37.42
17.44
17.30
29.77

Using the above unit prices, we were able to study the changes that occurred in Qi values
at the three holding cost rates. Table 4-11 shows the impact holding costs have on the allocation
of space within the Pyxis as items of greater costs are kept in lower quantities. These findings
demonstrate the relationship between holding costs and the optimal Qi values. In general, as the
holding cost rate used to determine the holding costs increases the resulting Qi values will
increase. Thus, there is a reduced penalty associated with carrying greater inventory. However,
it is important to recognize that Qi will not always increase with increases in r values, which is
due to the influence of demand characteristics (i.e. average daily demand and standard deviation
of daily demand) of individual items in setting the reorder points and ultimately calculating the
optimal order quantities.
Next, we summarize the observed differences in optimal order quantities across the three
levels of holding cost rate. Specifically, pairwise comparisons are made to show the overall
differences (%) that result when this rate changes. In Table 4-12, the following measures are
provided: average difference, absolute average difference, standard deviation of change,
minimum difference, and maximum difference.
Table 4-12: Pairwise Comparisons of Percent Differences for Varying Holding Cost
Qi1 and Qi2

Qi2 and Qi3

Qi1 and Qi3

Average Percent Difference

2.46%

3.34%

5.15%

Absolute Avg. Difference

6.92%

5.51%

11.66%

Standard Deviation of Change

6.61%

4.64%

10.20%

Minimum Difference

0.03%

0.15%

0.06%

Maximum Difference

24.86%

16.72%

37.42%

These results report the differences in optimal ordering values observed as we move from
one level of holding cost to another. Since some Qi values will increase and others will
decreases as the holding cost rate is modified, it is useful to look at the overall average change
81

that occurs with shifts in rate. One can see from Table 4-12 that there is a 2.46% average
increase in Qi values across all items when the holding cost rate shifts from r1 (0.001) to r2
(0.002) with a slightly greater increase of 3.34% when the value changes from r2 to r3 (0.003).
As expected, this difference is even greater when the disparity in rates is larger, which is
evidenced by the r1 to r3 comparison that yielded a 5.15% increase in Qi values. With respect to
the absolute average difference, we see that the difference in optimal order quantities is 6.92%
and 5.51% for the first pair-wise comparisons with the change being almost 12% when the rate
increased from r1 to r3. The move from r2 to r3 yielded the smallest standard deviation in change,
which was followed by the r1 to r2 and finally r1 to r3. The minimum percent changes were nearly
the same for each pairwise comparison with each having items that exhibited less than 0.15%
difference in Qi values at the specified rates; however, this was not true for the maximum
differences in Qi. When the rate changes from r2 to r3, the largest maximum change value occur
was 16.72%, which was the lowest observed maximum change across the comparisons. Again,
this was followed by the shift from r1 to r2 at 24.86% and finally r1 to r3 at 37.42%.
4.2.3 Strategic Decision Support
This research focused on healthcare SCM issues as they related to pharmaceutical
inventory management within the hospital; however, there are a number of additional
stakeholders interested in the product formulary. As previously established in the beginning of
this work, these stakeholders all have their own objectives, which are often at odds. At the heart
of this issue is tradeoff between product variety and economies of scale. Physicians value their
prescribing autonomy and push pharmacy directors and hospital administrators to offer a wider
selection of drugs. Pharmacy directors are constantly negotiating with the GPO over this same
issue. According to pharmacy managers at the participating hospital, efforts are made to
accommodate doctors and to offer greater variety, but product and operational costs are also a
82

concern. Unfortunately, the cost impacts of changes in the product formulary are not
documented and remain essentially unknown to these parties. We estimate the influence of these
costs using our models to provide valuable information for the hospital when negotiating with
doctors or the GPO on issues related to the formulary and pharmaceutical inventory
management.
A significant benefit of using this decision support tool is the allowance of quick
evaluations of the managerial tradeoffs, which facilitates strategic decision making in the
pharmacy. When suggestions are made that impact the number of workers or work shifts
available in the pharmacy or when changes to the formulary are discussed, the pharmacy director
can quickly analyze the issues based on our models and computation and identify the resulting
changes in service levels, workloads, and operating costs.

Analysis and Interpretation of the Tradeoffs for the Hospital Inventory Management
Besides the simple and fast computation, the practical importance of the simplified

approximate Model 3 is the straightforward and efficient way of showing the tradeoffs in
pharmaceutical supply management. It provides a simple and efficient strategic decision support
tool to analyze the tradeoffs among the three key performance indicators:

the service level (emergency refill workload),

the available space (depending on the variety of drugs formulary), and
the number of orders (refill workload) per day.
Next we summarize the most important aspects of the tradeoffs.

The refill workload of a day for a Pyxis MedStation is

N = the expected number of total orders (refills) per day.
The optimal N can be approximated by
N* = (di /Qi) = W2 / M.

(38)

83

The tradeoff between the number of orders and free storage space for a given service
level can be expressed in the simple form
N M = W2.

(39)

In this expression W = (vi di) is fixed, it depends only on the number of drugs (formulary)
and on the volume and demand of drugs. The parameter M = M - [vi (si - ui)] is the remaining
free storage space for cycle stock. Since si depends on the service level, implicitly equation (39)
also includes the service level tradeoff. M decreases with an increased service level requirement.

Managerial Tradeoffs: What-if and Sensitivity Analysis

This section of the chapter identifies several key managerial tradeoffs and demonstrates

the utility of this research in modeling the critical issues for management. Specifically, we
examine the impacts of changing formulary, available space, and worker capacity on our primary
performance measures. For a fixed service level the tradeoff curve has a hyperbola shape that is
shifted to the right (higher space requirement) with increasing service level requirement. This
property allows a simple visual tradeoff analysis, and a simple example is illustrated next.
In this research we use the example of a particular Pyxis MedStation of the case
hospital for illustration. All transactions (demand, refill, inventory position) and control
parameters (si, Si) for two years are available in Excel files. The number and composition of
drugs (formulary) is changing from time to time. We use the same example as before for
illustration and selected 70 drugs with highest usage rate out of the 214 drugs. These items make
up 71% of the total usage and around 70% of the total volume of the Pyxis. For available
volume, M, we took away the space used by the remaining low volume items that were
disregarded. This sample was used to illustrate the relevant tradeoffs.

84

Figure 4-2: Performance Indicator Relationships and Tradeoffs

Figure 4-2 illustrates the relationship between available space for cycle stock, M, and
average daily refills, N, at a given service level. Consider what happens when the available space
increases at the 95% service level. As space availability increases, one can read the decrease in
the average number of daily refills from the graphs. For example, if we double the space
allocated for items from 300 to 600, the number of orders will decrease from 9 to 5.
Understanding this relationship between space and refills allows us to analyze what will happen
if we change service levels. As such, if we shift from 95% to 99% service level at the same space
utilization, the graph demonstrates the expected increase in the refill requirements. Consider
what happens at a fixed storage capacity level of 600. As the service level increases from 95% to
99%, the expected number of refills per day increases approximately by 25%.
The available space is directly connected to the formulary (the number of different drugs
provided). Even if the total demand for the drugs doesnt increase, a bigger number of items
85

require a larger amount of safety stocks. The safety stock increase is proportional to the
variability of demand. For lower level of demand, typically the / rate increases decreasing the
available free space for cycle stock. For estimating a more accurate effect of increased formulary
a more detailed comparison of the reorder points is required based on formula (32) and (33).

Changes to the Available Space and Product Formulary

As established early in this work, there are a number of important stakeholder conflicts

present throughout this supply network. The formulary is at the center of many of these
disagreements and represents a substantial area of cost for the organization. Next we investigate
the effects of adding or removing items from a Pyxis. The inclusion of extra items means that
less space is available at the item level for cycle stock in the Pyxis. As such, reducing the
number of items stored in the Pyxis will have the opposite effect as more space becomes
available for cycle stock for individual items in the machine.
In Table 4-13 a simple demonstration of the formulary changes and the resulting
influence on pharmacy workload is provided. Again, we start with the sample of 70 drugs found
in a particular Pyxis MedStation and modify the product mix to include or omit 20% of items
in the local storage unit. To establish a reasonable idea of the impact such changes to the
formulary will have on the number of expected refills (N), we consider both high usage and low
demanded items for the situation where items are added. In contrast, we only consider the low
usage items for circumstances where items are removed from the Pyxis, which is practical
given the small likelihood of high usage items being eliminated from the formulary. This allows
pharmacy directors to gain valuable insight into the cost implications resulting from introducing
new items to the local depot.

86

Table 4-13: Impact of Adding or Subtracting Items to Local Formulary

Changes in Formulary
Baseline (70 items)
20% Additional High Usage Items
20% Additional Low Usage Items
20% Fewer Items

Expected Number of Daily Refills (N)

3.58
5.31
4.08
2.70

%
Difference
--48.32%
13.97%
-24.58%

As shown above in Table 4-13, increasing or decreasing the number of items in the
Pyxis will influence the numbers of expected daily refills needed for the machine. It is
important to recognize the impact these scenarios will have on the operational, tactical, and
strategic decisions facing managers. Another key issue is the relationship average daily demand
and demand types have on impacting this measure.
Regardless of the daily demand, adding items will raise the number of refills. First,
consider the situation where low usage or lumpy demand items are added to the local depot. In
this case one can observe approximately a 14% increase in the expected daily refills at the
Pyxis. This additional workload may seem relatively small; however, this is only for one
machine. Considering the participating hospital has over 85 Pyxis MedStations distributed
around the facility in various CUs, this increased workload poses significant problems from both
a cost and worker capacity perspective. Second, we examine the effect of adding high usage
items with daily demand values similar to the top 20% of items currently housed in the Pyxis
unit. If such products are added to the formulary, results indicate an increase in expected daily
refills of more than 48%. On the other hand, we noticed almost a 25% drop in refills when the
bottom 20% of drugs in the product formulary were removed. This demonstrates the importance
of both reasonably restricting the product formulary and evaluating the items in the Pyxis on a
frequent basis for possible reductions. Again, these results are significant for managers

87

attempting to understand the influence of product variety on pharmaceutical inventory control

and management.

Changes to Worker Capacity

In this context the number of refills that can be accomplished by pharmacists and

technicians during an 8-hour shift is relatively fixed. Pharmacists provide oversight of the
refilling process due to the sensitive nature of controlling pharmaceuticals and the need to verify
the prescribed medications before distributing these treatments to the patients. Pharmacy
technicians are responsible for preparing drug carts used in transporting items from the pharmacy
to the Care Units (CUs) and in the actual restocking of the Pyxis. As a result, any increases in
the number of refills required at the CUs can only be satisfied through incremental increases in
the number of work shifts. As the service level increases, one expects that smaller numbers of
items require refills on a daily basis to prevent expensive stockout occasions. On the other hand,
higher service level, and the resulting higher safety stock, takes away space from cycle stock
increasing the expected number of regular refills per day. Table 4-14 provides a simple
illustration of these relationships.
Table 4-14: Incremental Increases to Worker Capacity
Shifts
1
2

Refills
95% SL 99% SL
10
15
20
30

Knowing this relationship and the expected number of refills required at any given
Pyxis allows managers to better schedule human resources and ensure that the workload does
not exceed worker capacity. By optimizing the allocation of space within the Pyxis such that
the number of expected refills is minimized, the hospital pharmacy has the ability to keep
workloads at a controllable level and track changes over time. Control charts are useful in
88

monitoring workloads and in determining if the workloads are exceeding preset limits, which
indicates that conditions have changed and require attention. If the total workload for the system
is determined and the number of workers is known, managers can create a control chart using the
maximum, minimum, and average workloads to determine if the system is exceeding worker
capacity. If so, the hospital must add workers to satisfy that excess workload. As an example,
demand and the composition of items in the formulary can change over time. Trends indicating
workloads are approaching the pharmacy capacity limits indicate a need to reevaluate
pharmaceutical inventory control values and possibly adjust the number of technicians or work
shifts refilling the machines.
4.5 Summary of Findings
This chapter provided a description of the decision support tool, its design, and its utility.
In addition to this, numerous illustrations were provided demonstrating the decision support
capabilities of Model 2 and Model 3. With the consideration of holding and refill costs, Model 2
outperforms Model 3 in that it provides an allocation strategy which minimizes the sum of these
two costs. However, Model 3 is superior when holding costs are unknown or ignored, as they
are in our practical case. This model also excels in its simplicity and ease-of-use. Furthermore,
several simplified approaches to implementing these models on a daily basis were presented as
means of improving operational decisions when products are added to the formulary or when
demand characteristics change for an individual item. After an initial optimization run, these
techniques offer quick managerial decision support for estimating the close to optimal order
quantities and the min and max par levels for daily operations. Also, the included what-if
analyses provide invaluable information to administrators at the tactical and strategic levels of
decision making.

89

CHAPTER 5: CONCLUSION
This chapter provides closing discussions of the outcomes of this work. First, the
contributions of this research are identified. Then, the limitations are recognized along with
several possible extensions of the current project. Finally, concluding remarks are made
concerning the project as a whole.
5.1 Generalizability of Findings
This work is based on inventory control models that are commonly used in practice and
taught in academic settings. In addition, the case hospital, its dilemmas, and its current
management practices are believed to be commonplace amongst other such healthcare facilities.
The basic inventory problem considered in this research addressed the management desire to
reduce workload (emergency and daily refilling occasions) while considering a high service level
requirements and limited space availability, which are common inventory control constraints and
goals of managers regardless of industry. Models 2 and 3 are adaptable to any such
environments. Although the product characteristics make pharmaceuticals interesting to study,
the methods employed in this work are easily applied to any other product where similar
management constraints are present. For example, consider grocery stores that have large
numbers of items to consider with varying product characteristics, limited storage capacity for
these items, and substantial restocking activities that require considerable efforts. Here the
workload requirements and costs of emergency and daily refilling may be major concerns for
management, possibly even above inventory holding cost. In addition, this research allows for
an easy interpretation of the operational, tactical, and strategic tradeoffs when inventory holding
cost is also important and an optimal balance between holding and refill costs are sought. As
such, the results of this research are highly generalizable, and the situations in which Models 2
and 3 can be applied are nearly limitless.
90

5.2 Contributions
This project resulted in a number of theoretical and practical contributions that have
furthered the existing body of knowledge in healthcare supply chain management. The primary
goal of this work was to improve the current pharmacy inventory management policy and to
offer managerial support with the developed decision support tool. This has the potential of
being an important step toward improving the healthcare supply chain.
5.2.1 Contributions - Theoretical
This dissertation provides several key theoretical contributions. Three quantitative
models are presented that allow for the optimal setting of inventory control parameters (s, S)
given a multi-item, multi-location inventory management environment under multiple constraints
(i.e. service level, available space, and workload). Previous research in this area handle several
aspects of this research case; however, they fail to jointly consider the challenge of multi-item
and joint constraint with demand uncertainty or are too complex, time and data intensive for our
practical application. This project addresses those shortcomings and accomplishes the following:
1. provides an extensive review of the relevant managerial and quantitative literature as it
currently exists,
2. presents three quantitative models for determining the optimal min and max par values
(reorder point and order up to level) and allocating space to multiple items under multiple
constraints,
3. proposes two models (Models 2 and 3) and iterative procedures that both satisfy the
aforementioned constraints and offer simplified, practical alternatives to more
complicated optimization approaches,
4. applies these models and demonstrates their utility and performance using actual
pharmaceutical usage data, and
91

5. supports the inclusion of key demand characteristics (i.e. rate of daily demand variance to
average daily usage, holding cost rate, etc.) in the optimization process.
The quantitative models and Excel spreadsheets prepared are providing further
managerial support for pharmacists to analyze tradeoffs between service level and cycle stock,
test implications on cost savings and effects of parameters, and test a variety of parameter
settings. Furthermore, the tools facilitate the management of worker capacity. Based on refill
capacity requirements, pharmacy directors can manage the addition of pharmacists or overtime
and the addition of another shift of pharmacy technicians responsible for filling the Pyxis
MedStations around the hospital.
5.2.2 Contributions Practical
This work is of utmost practical relevance as it addresses real-world dilemmas and
concerns of hospital administrators and pharmacy managers. The magnitude of the healthcare
industry, ever-increasing healthcare costs, and role of pharmaceuticals within this industry
demonstrate the importance of continued research in this area. In addition, the stakeholders,
products, and policies create a unique research opportunity, and this investigation provides
detailed study of these issues. Specifically, this dissertation offers the following contributions:
1. It achieves the primary objective of the project and offers improvements to the
current inventory management practices at the local hospital;
2. It provides a decision support tool that allows for optimization of pharmaceutical
inventory control at the local depots and that addresses managerial concerns at the
operational, tactical, and strategic levels of decision making;
3. It illustrates the managerial tradeoffs associated with changes in formulary, service
level, and space availability, as well as, offers quick, simple techniques for estimating

92

optimal min and max par levels on a daily basis when such changes occur and
complete optimization of the system is impractical;
4. It shows the influence of holding and refill costs on optimizing inventory control
parameters and demonstrates the benefit of including such costs in inventory policy
comparisons.
Furthermore, for the operational inventory decision, the approaches offered provide the
min and max par levels that control the automated ordering system. These parameters are based
on near-optimal allocation policies of cycle stock and safety stock under storage space constraint.
As proved, the suggested selection of the control parameters si and Si

provide consistent service level,

allocate the safety stock to decrease the workload of emergency refilling, and

allocate the cycle stock space decreasing the workload of daily refilling.

For the tactical and strategic decisions, the presented Models 2 and 3 can be applied as a
simple visual decision support tool to analyze the tradeoffs among the refill workload, the
emergency workload, and the variety of drugs offered. This work illuminates the relationship of
these tradeoffs to the three key performance indicators at a local care unit: the expected number
of daily refills, the service level, and the storage space utilization.
5.3 Limitations
As with any complicated area of study, one must consider the limitations of the
investigation. Here it was necessary to restrict the scope of the project to allow for appropriate
comparisons. Within pharmaceutical products one can easily observe great variety in the type,
shape, and purpose of individual drugs. In addition, these variations often necessitate special
handling and storage consideration both in the central pharmacy and in the local depots. To
facilitate the evaluation of the research models and comparisons amongst allocation strategies,
93

this study concentrated on non-controlled substances that were readily available for examination
and that could be stored using a consistent layout within a Pyxis unit. All of the items in the
sample could be stored using a matrix layout that uses dividers to separate drugs in the machines
drawers. Since the hospital has discretion in the setup of the Pyxis, this is not a major
limitation of this study and is a reasonable approach to analyzing this research problem.
Further study is necessary to examine other drug types, specifically controlled
substances, and the various storage restrictions and requirements of these items. There are a
number of drawer layouts and storage options available to the pharmacy, and the greatest
variation in storage methods is employed with these controlled substances. Unfortunately, this
may prove very difficult given the legal and ethical guidelines for handling such drugs. In
addition, the specific issue of perishability was largely ignored for this study; however, this issue
is somewhat covered with the impact of formulary changes on optimal space allocation within
the Pyxis.
Finally, the absence of a demonstration of Model 1 with optimal solution is a limitation
of this study. The solution of Model 1 using approximations (4) to (7) is more straightforward,
and it could be applied to check different practical scenarios in this research case. However, the
managerial problem is that the ordering and shortage cost factors are very difficult to provide.
As stated previously, the hospital pharmacy was unable to produce any specific cost data to
support this research activity and does not include these costs in current policies. The technical
problem is that the large number of items (n = 250 to 300 per local depot), the nonlinearity, and
the stochastic demand make the solution challenging and time consuming. Given the pharmacy
objectives and the practical requirements, this was not a feasible solution for this case.
However, to address this limitation, actual usage data was used to construct a simulation
of daily transactions at the local depot for all 70 items included in the sample used for analytical
94

purposes. Using the min and max par levels for the GPO suggested policy (GPO), the hospital
modified (HM) inventory policy, and our Model 3 (OM) policy and the daily transaction data for
one year, this simulation tested the accuracy of model predictions by showing the reduction of
inventory with use, indicating the amount of shortages (stockouts) when they occurred, and the
amount and timing of refill occasions. In addition, the simulation calculated the average
inventory levels for each item given the three policies. As evidenced in Chapter 3, the
differences between model predictions and simulation results are very low and support the
accuracy of this approach.
5.4 Future Research Directions
There are plenty of extension possibilities. Some of them are technical, quantitative in
nature, others are managerial extensions. The main technical improvements include examining
the effect of special demand types and the consideration of multiple objectives in which Multiple
Criteria Decision Making (MCDM) tools may be applied. Specifically, more research is
necessary to examine the effects of low volume demand, lumpy demand, and auto-correlated
demand when subsequent days have a high demand applying a treatment followed by a longer
period without demand.
Some of the most important managerial extensions include examining the effect of
demand uncertainties on workload. Specifically, it is important to determine the uncertainty of
the workload by analytic estimation or by simulation and to determine if there is enough worker
capacity with certain reliability. Then, the question becomes if capacity is not enough, how
many workers (or additional shifts) to add to satisfy that excess workload?
Another significant extension of this research is the opportunity to provide more specified
quantitative support for negotiations between pharmacy administrators and medical doctors by
analyzing the effect of extending formulary and the cost vs. product variety tradeoff. The current
95

research effort and these recommended extensions will provide further insight into stakeholder
preferences and acceptable tradeoffs.
5.5 Concluding Remarks
This dissertation was introduced with a real world scenario of an ER supply problem
that with variations in details will possibly be experienced at some time by individuals who read
this work. A broad view of characteristics that make healthcare supply chain management a very
intriguing area of study has been presented. The healthcare industry operates in a unique manner
in that it has a number of special constraints all being enforced at once. Specifically, the focus is
on maintaining a high level of service while restraining costs, as opposed to reducing them. In
traditional supply chains, these variables are manipulated to achieve the optimal economic
operating setting. However, in the healthcare industry, tradeoffs are not as easily achieved.
Furthermore, medicines are special products as noted by Almarsdttir and Traulsen (2005),
which make pharmaceutical inventory management within the hospital a prime topic for study.
The significance of the healthcare industry is demonstrated by the provided cost
breakdown and by conveying the importance of affordable, quality care to the described
stakeholders. As such, works like the current project are valuable from both the practical and
academic perspectives. The case is made for additional studies in this area, and an overview of
the current endeavor and its objectives is given. A review of current managerial and quantitative
modeling literature in this specific area is presented. It is important to recognize that the need for
continued research in this domain is warranted. Many of the proposed inventory control models
have yet to be tested in a healthcare setting or have not been evaluated under conditions of
varying demand types (i.e. intermittent or auto-correlated demand). Outsourcing and VMI in
some form are used extensively to alleviate strains on pharmacy resources and material
management issues; however, greater study is needed to truly understand the impact of such
96

activities throughout the supply network. Information technology plays a significant role in
determining the success of these solutions, and the case hospital applies advanced technology in
controlling medication throughout the hospital. The drugs are stored in the local depots (Pyxis
MedStations) in 86 different areas of the hospital with each area having a different selection of
drugs. The Pyxis MedStation registers each transaction date and quantity of demand
(withdrawal) and each delivery (refill) and the actual inventory level. It is connected to the
central depot (Tallyst system) through a computer network. It seems clear that investments in
these resources will continue in the future as hospitals increase their reliance on the information
created and controlled by these machines.
The findings presented within this dissertation have a profound reach and have the ability
to impact anyone connected to healthcare. These results suggest alternative methods for
allocating space for inventory within the local depots around the hospital and provide needed
tools for managerial decision making at all levels. The proposed models deliver simplified,
quick quantitative approaches that allow for setting the optimal reorder points and order up to
levels for items in a Pyxis unit, as well as, provide managers a mechanism for demonstrating
key managerial tradeoffs and negotiating with other stakeholders in the supply network.

97

REFERENCES
Allen, S. 1958. Redistribution of total stock over several user locations. Naval Research
Logistics Quarterly 5:5159.
Almarsdttir, A., and Traulsen, J. 2005. Cost-containment as part of pharmaceutical policy.
Pharmacy World and Science 27:144-148.
Andersson, J., and Melchiors, P. 2001. A two-echelon inventory model with lost sales.
International Journal of Production Economics 69:307315.
Arthur, J., and Ravindran, A. 1978. An efficient goal programming algorithm using constraint
partitioning and variable elimination. Management Science 24:867-868.
Arthur, J., and Ravindran, A. 1980. A branch and bound algorithm with constraint partitioning
for integer goal programming problems. European Journal of Operational Research 4:421-425.
Axster, S. 2003. Approximate optimization of a two-level distribution inventory system.
International Journal of Production Economics 81-82:545-553.
Banerjea-Brodeur, M., Cordeau, J-F., Laporte, G., and Lasry, A. 1998. Scheduling linen
deliveries in a large hospital. Journal of the Operational Research Society 49:777780.
Beyer, D., Sethi, S.P., and Sridhar, R. 2001. Stochastic multiproduct inventory models with
limited storage. Journal of Optimization Theory and Applications 111:553588.
Bolton, C., and Gordon, J. 1991. Health care material management, Working paper, Queens
University, Kingston.
Bower, M., and Garda, R.A. 1985. The role of marketing in management. McKinsey Quarterly,
Autumn, 34-46.
Burns, L., and Wharton School Colleagues. 2002. The Health Care Value Chain Producers,
Purchasers, and Providers. San Francisco: Jossey-Bass.
Castles, F. 2004. The Future of the Welfare State: Crisis Myths and Crisis Realities. Oxford:
Oxford University Press.
Catlin, A., Cowan, C., Heffler, S., and the National Health Expenditure Accounts Team. 2007.
National Health Spending in 2005: The Slowdown Continues. Health Affairs 26:142-153.
Centers for Medicare and Medicaid Services, Office of the Actuary, National Health Statistics
Group. 2007. 2005 National Health Care Expenditures Data. http://www.cms.hhs.gov/
nationalhealthexpenddata/01_overview.asp? (accessed October 20, 2007).
Chiu, H. 1995. An approximation to the continuous review inventory model with perishable
items and lead times. European Journal of Operational Research 87:95-108.
98

Clark, A., and Scarf, H. 1963. Optimal policies for a multi-echelon inventory problem.
Management Science 6:475490.
Comas-Herrera, A. 1999. Is there convergence in the health expenditures of the EU Member
States? In Health Care and Cost Containment in the European Union, ed. E. Mossialos and J.
Grand, 197-218. Aldershot: Ashgate.
Croston, J. 1972. Forecasting and stock control for intermittent demands. Operational Research
Quarterly 23:289304
Culyer, A. 1990. Cost containment in Europe. In Health Care Systems in Transition. The Search
for Efficiency, ed. OECD, 29-40. Paris: OECD.
Das, C. 1975. Supply and redistribution rules for two-location inventory systems: One period
analysis. Management Science 21:765776.
Dellaert, N., and van de Poel, E. 1996. Global inventory control in an academic hospital.
International Journal of Production Economics 4647:277284
Deuermeyer, B., and Schwarz, L. 1981. A model for the analysis of system service level in a
warehouse-retailer distribution system. In Multi-Level Production/Inventory Control Systems:
Theory and Practice, ed. L. Schwartz, 163-193. North-Holland: Studies in Management Science.
Ehrhardt, R. 1979. Power Approximation for Computing (s, S) Inventory Policies. Management
Science 25:777-786.
Eppen, G., and Schrage, L. 1981. Centralized ordering policies in a multi-warehouse system with
lead times and random demand. In Multi-Level Production/Inventory Control Systems: Theory
and Practice, ed. L. Schwartz, 51-67. North-Holland: Studies in Management Science.
Evans, J., and Berman, B. 2001. Conceptualizing and operationalizing the business-to-business
value chain. Industrial Marketing Management 30:135-48.
Fries, B. 1975. Optimal order policy for a perishable commodity with fixed lifetime. Operations
Research 23:46-61.
Gallego, G., Queyranne, M., and Simchi-Levi, D. 1996. Single-Resource Multi-Item Inventory
Systems. Operations Research 44:580595.
Gder, F., and Zydiak, J. L. 1999. Ordering policies for multi-item inventory systems subject to
multiple resource constraints. Computers and Operations Research 26:583597.
Guillen, A., and Cabiedes, L. 2003. Reforming pharmaceutical policies in the European Union:
A "penguin effect"? International Journal of Health Services 33:1-28.
Hariga, M., and Jackson, P. 1996. The Warehouse Scheduling Problem: Formulation and
Algorithms. Institute of Industrial Engineering Transactions 28:115-128.
99

Heffler, S., Smith, S., Keehan, S., Borger, C., Clemens, M., and Truffer, C. 2005. U.S. health
spending projections for 2004-2014. Health Affairs. http://content.healthaffairs.org/cgi/
content/abstract/hlthaff.w5.74#otherarticles (accessed on October 14, 2007).
Henning, W. 1980. Utilizing suppliers to the hospitals best interest. Hospital Material
Management Quarterly 1:39-47.
Hill, R., Seifbarghy, M., and Smith, D. 2007. A two-echelon inventory model with lost sales.
European Journal of Operational Research 181:753-766.
Hoadley, B., and Heyman, D. 1977. A two-echelon inventory model with purchases, depositions,
shipments, returns, and transshipments. Naval Research Logistics Quarterly 24:120.
Hollier, R., Mak, K., and Yiu, K. 2005. Optimal inventory control of lumpy demand items using
(s, S) policies with a maximum issue quantity restriction and opportunistic replenishments.
International Journal of Production Research 43:4929-4944.
Holmgren, H., and Wentz, J. 1982. Material Management and Purchasing for the Health Care
Facility. Washington, DC: AUPHA Press.
Iglehart, D. 1963. Dynamic Programming and Stationary Analysis of Inventory Problems in
Multistage Inventory Models and Techniques, H. E. Scarf, D. M. Gilford, and M. W. Shelly
(eds.), Stanford University Press, Stanford, California.
Ignall, E. 1966. Optimal Policies for Two-Product Inventory Systems, with and without Setup
Costs, PhD diss., Cornell University.
Ignall, E., and Veinott, A. 1969. Optimality of Myopic Inventory Policies for Several Substitute
Products. Management Science 18:284304.
International Society on Multiple Criteria Decision Making. 2008.
http://www.mit.jyu.fi/MCDM/intro.html (accessed on October 1, 2008).
Jarrett, P. 1998. Logistics in the health care industry. International Journal of Physical
Distribution and Logistics Management 28:741742.
Jayaraman, V., Burnett, C., and Frank, D. 2000. Separating inventory flows in the materials
management department of Hancock Medical Center, Interfaces 30:5664.
Jnsson, B., and Musgrove, P. 1997. Government financing of health care. In Innovations in
Health Care Financing, ed. G. J. Schieber, 41-64. Washington, DC: World Bank.
Kim, D. 2005. An integrated supply chain management system: A case study in healthcare
sector. Lecture Notes in Computer Science 3590:218-227.
Krishnan, K., and Rao, V. 1965. Inventory control in N warehouses. Journal of Industrial
Engineering 16:212215.
100

Kuriger, G., and Ravindran, A. 2005. Intelligent search methods for nonlinear goal programs.
INFOR 43:79-92.
Landry, S., and Beaulieu, M. 2000. tude internationale des meilleures pratiques de logistique
hospitalire. Montreal: Cahier de recherche du groupe CHAINE.
Lapierre, S., and Ruiz, A. 2007. Scheduling logistic activities to improve hospital supply
systems. Computers & Operations Research 34:624-641.
Lee, H., and Wu, J. 2006. A study on inventory replenishment policies in a two-echelon supply
chain system. Computers and Industrial Engineering 51:257-263.
Li, L., and Benton, W. 1996. Performance measurement criteria in health care organizations:
Review and future research directions. European Journal of Operational Research 93:449468.
Lian, Z., and Liu, L. 2001. Continuous review perishable inventory systems: models and
heuristics. IIE Transactions 33:809-822.
Liu, L., and Lian, Z. 1999. (s, S) continuous review models for products with fixed lifetimes.
Operations Research 47:150-158.
Lunn, T. 2000. Ways to reduce inventory. Hospital Material Management Quarterly 21:17.
Marmor, T., and Okma, K. 1998. Cautionary lessons from the West: what (not) to learn from
other countries experience in the financing and delivery of health care. In The State of Social
Welfare, 1997. International Studies on Social Insurance and Retirement, Employment, Family
Policy and Health Care, ed. P. Flora, P. R. de Jong, J. Le Grand, and J. Y. Kim, 327-350.
Aldershot: Ashgate.
Meijboom, B., and Obel, B. 2007. Tactical coordination in a multi-location and multi-stage
operations structure: A model and a pharmaceutical company case. Omega 35:258-273.
Michelon, P., DibCruz, M., and Gascon, V. 1994. Using the tabu search method for the
distribution of supplies in a hospital. Annals of Operations Research 50:427435.
Minner, S., and Silver, E. 2005. Multi-product batch replenishment strategies under stochastic
demand and a joint capacity constraint. IIE Transactions on Scheduling and Logistics 37:469
479.
Minner, S., and Silver, E. 2007. Replenishment policies for multiple products with compoundPoisson demand that share a common warehouse. International Journal of Production
Economics 108:388-398.
Naddor, E. 1980. An analytic comparison of two approximately optimal (s, S) inventory policies.
Technical Report 330. Baltimore: Department of Mathematical Science, Johns Hopkins
University.
101

Nahmias, S. 1975 Optimal ordering policies for perishable inventory - II. Operations Research
23:735-749.
Nahmias, S. 1982. Perishable inventory theory: a review. Operations Research 30:680-707.
Nahmias, S., and Smith, S. 1994. Optimizing inventory levels in a two-echelon retailer system
with partial lost sales. Management Science 40:582596.
Nicholson, L., Vakharia, A.J., and Selcuk Erenguc, S. 2004. Outsourcing inventory management
decisions in healthcare: Models and application, European Journal of Operational
Research 154:271-290.
OECD. 1994. Health: Quality and Choice. Paris: OECD, OECD Health Policy Studies No. 4.
OECD. 1995. New Directions in Health Care Policy. Paris: OECD, OECD Health Policy Studies
No. 7
OECD. 1996. Health Care Reform. The Will to Change. Paris: OECD, OECD Health Policy
Studies No. 8.
Ohno, K., and Ishigaki, T. 2001. A multi-item continuous review inventory system with
compound Poisson demands. Mathematical Methods of Operations Research 58:147165.
Perini, V., and Vermeulen Jr., L. 1994. Comparison of automated medication-management
systems. American Journal of Hospital Pharmacy 15:1883-1891.
Pitta, D., and Laric, M. 2004. Value chains in health care. Journal of Consumer Marketing
21:451-464.
Porter, M. 1985. Competitive Advantage: Creating and Sustaining Superior Performance. New
York: The Free Press.
Prashant, N. 1991. A systematic approach to optimization of inventory management functions.
Hospital Material Management Quarterly 12:3438.
Prosser, H., and Walley, T. 2005. A qualitative study of GPs and PCO stakeholders views on
the importance and influence of cost on prescribing. Social Science & Medicine 60:13351346.
Prosser, H., and Walley, T. 2007. Perceptions of the impact of primary care organizations on GP
prescribing: The iron fist in the velvet glove? Journal of Health, Organisation and Management
21:5-26.
Rivard-Royer, H., Landry, S., and Beaulieu, M. 2002. Hybrid stockless: A case study.
International Journal of Operations and Production Management 22:412424.
Roberts, D. 1962. Approximations to optimal policies in a dynamic inventory model studies. In
Applied Probability and Management Science, ed. K. Arrow, S. Kardin, and H. Scarf, 207-229.
Stanford: Stanford University Press.
102

Rogers, D., and Tsubakitani, S. 1991. Inventory positioning/partitioning for backorders

optimization for a class of multi-echelon inventory problems. Decision Sciences 22:536-558.
Rothgang, H., Cacace, M., Grimmeisen, S., and Wendt, C. 2005. The changing role of the state
in healthcare systems. European Review 13 (supplement):187-212.
Saber, H., and Ravindran, A. 1993. Nonlinear goal programming: theory and practice: a survey.
Computers and Operations Research 20:275-291.
Saber, H., and Ravindran, A. 1996. A partitioning gradient based (PGB) algorithm for solving
nonlinear goal programming problems. Computers and Operations Research 23:141-152.
Sani, B., and Kingsman, B. 1997. Selecting the best periodic inventory control and demand
forecasting methods for low demand items. Journal of the Operational Research Society 48:700713.
Scarf, H. E. 1960. The Optimality of (s, S) Policies in the Dynamic Inventory Problem in
Mathematical Methods in the Social Sciences, Stanford University Press, Stanford, California.
Schmidt, C., and Nahmias, S. 1985. (S 1, S) policies for perishable inventory. Management
Science 31:719-728.
Schneider, H. 1978. Methods for determining the reorder point of an (s, S) ordering policy when
a service-level is specified. Journal of the Operational Research Society 29(12) 1181-1193.
Schneider, H. 1981. Effect of service-levels on order-points or order-levels in inventory models.
International Journal of Production Research 19:615-631.
Schneider, H., D. B. Rinks. 1989. Optimal policy surfaces for a multi-item inventory problem.
European Journal of Operational Research 39(2) 180-191.
Schneider, H., D. B. Rinks, P. Kelle. 1995. Power approximations for a two-echelon inventory
system using service levels. Production and Operations Management 4(4) 381400.
Seo, Y., Jung, S., and Hahm, J. 2001. Optimal reorder decision utilizing centralized stock
information in a two-echelon distribution system. Computers and Operations Research 29:171193.
Simpson Jr., K. 1959. A theory of allocations of stocks to warehouses. Operations Research
7:797805.
Sinha, D., and Matta, K. 1991. Multiechelon (R, S) inventory model. Decision Sciences 22:484
499.
Smith-Daniels, V. 2006. Supply Chains Critical to Well-Being of Health-Care Systems.
Knowledge@W.P. Carey. http://knowledge.wpcarey.asu.edu/article.cfm?articleid=1245
(accessed on October 14, 2008).
103

Srinivasan, V., and Shocker, A. 1973. Estimating the weights for multiple attributes in a
composite criterion using pairwise judgments. Psychometrika 38:473-492.
Syntetos, A., and Boylan, J. 2006. On the stock control performance of intermittent demand
estimators. International Journal of Production Economics 103:36-47.
Tarabusi, C., and Vickery, G. 1998. Globalization in the pharmaceutical industry, part 1 and 2,
International Journal of Health Services 28(1):67105.
Tarabusi, C., and Vickery, G. 1998. Globalization in the pharmaceutical industry, part 1 and 2,
International Journal of Health Services 28(2):281303.
The Plunkett Research Group. 2008. Health Care Trends. http://www.plunkettresearch.com/
Industries/HealthCare/HealthCareTrends/tabid/294/Default.aspx (accessed September 10, 2008).
van Merode, G., Groothuis, S., and Hasman, A. 2004. Enterprise resource planning for hospitals.
International Journal of Medical Informatics 73:493-501.
Veinott, A. 1965. Optimal Policy for a Multiproduct, Dynamic Nonstationary Inventory
Problem. Management Science 12:206222.
Veral, E., and Rosen, H. 2001. Can a focus on costs increase costs? Hospital Material
Management Quarterly 22:2835.
Vissers, J., Bertrand, J., and de Vries, G. 2001. A framework for production control in health
care organizations. Production Planning Control 12:591604.
Wagner, H., M. O'Hagan, B. Lundh. 1965. An empirical study of exactly and approximately
optimal inventory policies. Management Science 11:690-723.
Weiss, H. 1980. Optimal ordering policies for continuous review perishable inventory models.
Operations Research 28:365-374.
Yu, P., and Zeleny, M. 1975. The Set of All Nondominated Solutions in Linear Cases and a
Multicriteria Simplex Method. Journal of Mathematical Analysis and Applications 49:430-468.
Zhu, L., Wang, H., and Zhao, L. 2005. Comparative research on three-echelon and two-echelon
medicine inventory model with positive lead-time. Journal of Southeast University (English
Edition) 21:500-505.
Ziegler, H. 1982. Solving certain singly constrained convex optimization problems in production
planning. Operations Research Letters 1:246-252.

104

APPENDIX A: PYXIS MEDSTATION AND MATRIX DRAWER

Pyxis MedStation 3500

6-drawer main*
22.8 W x 27 D x 55 H

105

Example of a Pyxis matrix drawer that uses plastic dividers to create a variety of storage
configurations. This particular matrix is setup to house 40 different products; however, the
maximum number of separate cubicles is 48. Dividers can be added/removed to accommodate
different numbers of items with varying sizes and shapes.

106

APPENDIX B: POWER APPROXIMATION

For the Power Approximation formula (11) the rational function p(yi) is defined as

p(yi ) =

with notation

a 0 +a 1 w+a 2 w 2 +a 3 w 3

b 0 +b 1 w+b 2 w 2 +b 3 w 3 +b 4 w 4

w = ln (25/yi2 )
and with given constants:
a0 = -5.3925569
a1 = 5.6211054
a2 = -3.8836830
a3 = 1.0897299
b0 = 1.0000
b1 = -0.72496485
b2 = 0.507326622
b3 = 0.0669136868
b4 = -0.00329129114

107

APPENDIX C: MODEL 2 CALCULATIONS FOR REORDER POINT AND OPTIMAL Q

Step 0:

Formulae:
hi = rci

K
r

M1 = M' - [vi(si(1)-ui)]
Q' = (2diK/hi)

M1

100
0.002
4860.14

V' = viQ'i
Step 1:

V'

Q"I = (M1/V)*Q'i
i = diK/vi(Q'i)2 - hi/2vi

V = viQi

17874.59

V0

V1

V2

V3

V4

V5

V6

V7

V8

V9

V10

V11

757.39

1063.28

1484.05

2052.61

2801.62

3753.19

4898.67

4207.27

4511.35

4692.29

4791.93

4844.36

-2260.01%

28.77%

28.35%

27.70%

26.73%

25.35%

23.38%

-16.43%

6.74%

3.86%

2.08%

1.08%

' = min i ; " = max i

Step 2:

= (' + ")/2

Step 3:

Qi() = [2diK/(hi/2 + vi)

Step 2

V' = viQi()

Drug
ZOFR2
ACET3UD
SENOKOTS
H100U10ML
PROM25I
KDUR20
DIPH50I
DEX4I
NEXI40C
NEUR300
NYSTSUSP5ML
ZOFRT
R10V
SODB650T
SUCRALUD
DIG125UD
Sum

"

Values

0.005

5.980

2.993

F(Q)

Iteration 1

0.005

2.993

1.499

F(Q )

517.97

775.42

775.42

775.42

775.42

775.42

775.42

515.10

515.12

515.12

515.12

515.12

515.12

200.86%

117.21%

59.49%

20.24%

-5.92%

-22.91%

16.06%

8.43%

4.34%

2.20%

1.11%

0.55%

diK
1663.49
1053.95
829.16
760.22
528.07
476.29
412.53
398.91
384.47
373.30
300.27
296.73
265.40
68.39
67.30
66.21

hi/2
0.01
0.02
0.01
0.03
0.03
0.00
0.02
0.01
0.01
0.11
0.11
0.02
0.01
0.09
0.09
0.02

vi

viQ()i
242.61
121.37
118.55
252.43
106.10
130.94
97.94
97.08
115.66
47.12
73.60
81.58
94.66
37.56
37.26
30.37
4844.36

Q
121.30
121.37
118.55
50.49
70.74
65.47
65.29
64.72
57.83
47.12
36.80
54.38
47.33
18.78
18.63
30.37

QL
122.89
122.62
120.03
51.11
71.48
66.34
66.04
65.47
58.57
47.32
37.04
54.98
47.92
18.92
18.77
30.68

F(QU)
14.43
10.86
7.64
16.59
9.25
7.53
7.39
7.13
7.10
13.32
12.38
6.54
6.14
5.27
5.23
2.73
517.97

0.752

0.379

Difference

0.005

0.379

0.192

16.63
10.54
8.29
7.60
5.28
4.76
4.13
3.99
3.84
3.73
3.00
2.97
2.65
0.68
0.67
0.66

Sigma
13.10
8.50
6.97
5.54
4.81
4.30
3.88
3.95
5.36
4.60
4.15
3.14
2.92
1.67
1.49
0.85

Size (vi)

Price ci

2
1
1
5
1.5
2
1.5
1.5
2
1
2
1.5
2
2
2
1

100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100

6
18
5
30
25
4
16
15
8
115
115
20
11
87
87
18

hi
0.0118
0.0359
0.0109
0.0608
0.0503
0.0079
0.0328
0.0298
0.0156
0.2292
0.2292
0.0399
0.0227
0.1736
0.1736
0.0365

Q'I = QU
530.86
242.19
390.93
158.18
144.83
346.63
158.57
163.73
221.84
57.08
51.19
121.91
153.06
28.07
27.84
60.27

Iteration 11

520.82

0.005

Average

Iteration 10

526.45

Iteration 4

0.099

Iteration 9

537.45

Iteration 3

0.052
0.075
0.064
0.058
0.055
0.054

Iteration 8

558.56

F(QU) = [K(di/QU) + hi(QU/2)]

0.192

Iteration 7

597.82

If V > M, then '= and Q =Qi()

0.099
0.099
0.075
0.064
0.058
0.055

Iteration 6

597.80

F(Q )

0.005

Iteration 5

729.55

0.752

0.005
0.052
0.052
0.052
0.052
0.052

Iteration 4

932.38

1.499

Iteration 5

Iteration 3

1236.70

0.005

Iteration 6
Iteration 7
Iteration 8
Iteration 9
Iteration 10
Iteration 11

Iteration 2

1684.32

Iteration 2

F(QU) - F(QL) / F(QL) E

Iteration 1

2332.92

F(QL) = [K(di/QL) + hi(QL/2)]

Initial

If V < M, then "= and Q =Qi()

Step 4:

'

Initial

Step 0
viQ'i
1061.72
242.19
390.93
790.92
217.25
693.26
237.86
245.59
443.68
57.08
102.38
182.87
306.12
56.14
55.69
60.27
17874.59

Q"I = QL diK/vi(Q'i)
144.34
0.040
65.85
0.243
106.30
0.073
43.01
0.082
39.38
0.227
94.25
0.027
43.12
0.148
44.52
0.134
60.32
0.053
15.52
1.550
13.92
0.775
33.15
0.180
41.62
0.077
7.63
0.587
7.57
0.587
16.39
0.247

Step 1
hi/2vi
0.003
0.018
0.005
0.006
0.017
0.002
0.011
0.010
0.004
0.115
0.057
0.013
0.006
0.043
0.043
0.018

i
0.04
0.23
0.07
0.08
0.21
0.02
0.14
0.12
0.05
1.44
0.72
0.17
0.07
0.54
0.54
0.23

0.11
0.05
0.05
0.27
0.08
0.11
0.08
0.08
0.11
0.05
0.11
0.08
0.11
0.11
0.11
0.05

Step 3
Qi()
121.30
121.37
118.55
50.49
70.74
65.47
65.29
64.72
57.83
47.12
36.80
54.38
47.33
18.78
18.63
30.37

Step 4
U

Here we start with the inputs of average daily demand, standard deviation of daily demand (Sigma), the unit size (Vi), and the unit price (ci) for each item in the sample. To begin, we calculate the expected order quantity (Qi) using the Economic Order Quantity (EOQ) formula (13) and
use the notation of Qi'. Next, we determine the corresponding reorder point value (si) using the Power Approximation (11). If the space constraint (10) is fulfilled, we set the order up to level (Si) using (14). If the space constraint (10) is not fulfilled, we must decrease the Qi values to
(0)

(U)

satisfy this constraint. To accomplish this we must find the values of Qi that fit into the available space, M, defined by (16) using the initial values of si = si . We set Qi' as the upper bounds, Q , on the order quantities for the iteration and determine the overall volume needed for
these values (V') using (19). In the first iterative step we set the lower bounds, Q , according to (20) and calculate values for lambda () using (21). In the second iterative step we determine the overall value by averaging the min and max lambda values according to (22).
In the third step, we calculate a new Qi value, which is noted as Qi() using (23) and employ our decision rules. In the fourth step of the iterative process, calculate the functions for the upper and lower bounds according to (24) and (25) and employ the stopping rule (26).
(L)

If the relative difference is less than or equal to the preset acceptable error, which was 0.01 in this case, stop. Otherwise, continue the iterations beginning at step two and repeat until convergence. Using the optimal Qi values as determined by this process to recalculate
(2)

(3)

the reorder points (si) using (11). With the new si = si values, recalcuate M = M and solve sub problem 2A (17) again. The resulting Qi values may be different than those determined in the prior step in which case new si = si values are used for new iterations until the
difference between si and Qi is smaller than the preset accracy limit (E = 0.01). For futher explanation of this process, reference section 3.2.1 in the text.

108

F(QL)
14.26
10.80
7.56
16.43
9.19
7.44
7.33
7.07
7.02
13.31
12.35
6.49
6.08
5.26
5.22
2.72
515.12

APPENDIX D: MODEL 3 CALCULATIONS FOR REORDER POINT AND OPTIMAL Q

Tmin

Tmax
3

10

*M' = Hospital M'

Step 0: Initial Values
M' = Tmax (vidi)

5931.50

M0 = M' - [vi(si-ui)]

5357.58

W = (vidi)

131.97

V0 = M02 / 2W2

824.10

Iterative Process
Mk = M' - [vi(si-ui)]

Iteration 1
4904.59

Difference

Iteration 2
4859.79
0.009
% Change

M0

M1

M2

5357.58
---

4904.59
9.24%

4859.79
0.92%

Step 0
Drug
ZOFR2
ACET3UD
SENOKOTS
H100U10ML
PROM25I
KDUR20
OSCALD500
LEUCO5T
SODB650T
SUCRALUD
DIG125UD

Average
16.63
10.54
8.29
7.60
5.28
4.76
0.69
0.68
0.68
0.67
0.66

Sigma

Size (vi)

13.10
8.50
6.97
5.54
4.81
4.30
1.03
2.33
1.67
1.49
0.85

2
1
1
5
1.5
2
1.5
1
2
2
1

ui
13.48
8.70
7.07
5.82
4.83
4.32
1.11
4.33
2.37
1.98
0.87

si(0)
49.90
31.62
24.87
22.81
15.84
14.29
2.08
2.05
2.05
2.02
1.99

Qi

Step 1

(0)

si(1)

117.08
131.80
116.90
50.06
76.17
62.65
27.63
33.57
23.74
23.55
33.04

Qi

60.59
33.53
26.18
26.09
17.50
16.08
2.23
7.20
4.76
3.96
1.66

(1)

107.19
120.66
107.02
45.83
69.73
57.35
25.29
30.74
21.73
21.56
30.24

Step 2
si (2)
61.86
34.36
26.87
26.58
17.98
16.49
2.32
7.66
5.01
4.16
1.73

Qi(2)
106.21
119.55
106.04
45.41
69.10
56.83
25.06
30.45
21.53
21.36
29.97

Step 3
si

Di
92.73
110.86
98.97
39.59
64.27
52.51
23.96
26.13
19.16
19.38
29.09

61.86
34.36
26.87
26.58
17.98
16.49
2.32
7.66
5.01
4.16
1.73

Here we start with the inputs of average daily demand, standard deviation of daily demand (sigma), and the unit size (vi) for each item. We then calculate the undershoot quantity (ui)
(0)

(0)

using (3). To initialize the iteration, set the reorder point (si ) using (30) and then calculate the initial value of M0 according to (31). Next, calculate the order quantity (Qi )
using (34) where W is set according to (35). In the first iteration step adjust the reorder points and find the si values that provide the required service level, , for each item
(1)

(1)

using (11) and then find the new value of M1 using (44). Next, we use these new values to determine the corresponding Qi values according to (45). The iterations continue until
convergence of M is achieved. Once the final values of si are determined, set Si according to (46) where D is calculated by (47). For further explanation reference section 3.2.3 of the text.

109

Si
154.59
145.22
125.84
66.17
82.25
69.00
26.28
33.79
24.17
23.54
30.83

APPENDIX E: MODEL 2 CALCULATIONS FOR SENSITIVITY ANALYSIS

Step 0:

Step 1:

Formulae:

100

hi = rci

0.002

M1 = M' - [vi(si(1)-ui)]

Q' = (2diK/hi)

M'

4200.00

V' = viQ'i

M1

3080.84

Q"I = (M1/V)*Q'i

V'

27587.67

0.99
V = viQi

i = diK/vi(Q'i)2 - hi/2vi

Step 2

V'

V0

V1

V2

V3

V4

V5

V6

27587.67

2315.08

2890.78

3405.89

3116.85

2997.44

3055.39

3085.66

'

"

Values

' = min i ; " = max i

Initial

0.032

0.198

0.115

F(Q)

Step 2:

= (' + ")/2

Iteration 1

0.032

0.115

0.073

F(Q )

270.93

218.12

218.12

218.12

210.59

206.72

206.72

Step 3:

Qi() = [2diK/(hi/2 + vi)

Iteration 2

0.032

0.073

0.052

F(QL)

225.98

225.98

186.18

202.78

202.78

202.78

204.76

Difference

19.89%

-3.48%

17.15%

7.56%

3.85%

1.94%

0.96%

V' = viQi()

Iteration 3

0.052

0.073

0.063

If V < M, then "= and Q =Qi()

Iteration 4

0.063

0.073

0.068

If V > M, then '= and QL=Qi()

Iteration 5

0.063

0.068

0.065

F(QU) = [K(di/QU) + hi(QU/2)]

Iteration 6

0.063

0.065

0.064

Step 4:

Initial

Iteration 1

Iteration 2

Iteration 3

Iteration 4

Iteration 5

Iteration 6

F(QL) = [K(di/QL) + hi(QL/2)]

F(QU) - F(QL) / F(QL) E

Drug

Average

Sigma

Size (vi)

ui

hi

Price ci

EOQ

(1)

si

vi(si-ui)

Q'I = QU

Step 0
viQ'i

Q"I = QL

diK/vi(Q'i)2

Step 1
hi/2vi

diK

hi/2

Step 3
Qi()

vi

Step 4
viQ()i

Q = Qi(")

F(QU)

F(QL)

DRUG 1

10.00

10.00

10

100

0.0100

447.21

47.47

187.34

447.21

2236.07

68.08

0.043

0.001

0.04

1000.00

0.01

0.32

55.41

277.03

54.86

55.41

18.50

18.33

DRUG 2

10.00

10.00

10

100

0.0040

707.11

47.41

187.06

707.11

3535.53

107.65

0.017

0.000

0.02

1000.00

0.00

0.32

55.66

278.31

55.11

55.66

18.26

18.08

DRUG 3

10.00

10.00

10

100

0.0100

447.21

42.20

64.39

447.21

894.43

68.08

0.108

0.003

0.11

1000.00

0.01

0.13

86.61

173.22

85.78

86.61

12.09

11.98

DRUG 4

10.00

10.00

10

100

0.0040

707.11

42.05

64.11

707.11

1414.21

107.65

0.043

0.001

0.04

1000.00

0.00

0.13

87.60

175.21

86.74

87.60

11.70

11.59

DRUG 5

10.00

4.00

5.8

100

0.0100

447.21

25.35

97.73

447.21

2236.07

68.08

0.043

0.001

0.04

1000.00

0.01

0.32

55.41

277.03

54.86

55.41

18.50

18.33

DRUG 6

10.00

4.00

5.8

100

0.0040

707.11

25.33

97.65

707.11

3535.53

107.65

0.017

0.000

0.02

1000.00

0.00

0.32

55.66

278.31

55.11

55.66

18.26

18.08

DRUG 7

10.00

4.00

5.8

100

0.0100

447.21

23.78

35.96

447.21

894.43

68.08

0.108

0.003

0.11

1000.00

0.01

0.13

86.61

173.22

85.78

86.61

12.09

11.98

DRUG 8

10.00

4.00

5.8

100

0.0040

707.11

23.74

35.87

707.11

1414.21

107.65

0.043

0.001

0.04

1000.00

0.00

0.13

87.60

175.21

86.74

87.60

11.70

11.59

DRUG 9

5.00

5.00

100

0.0100

316.23

21.33

81.67

316.23

1581.14

48.14

0.043

0.001

0.04

500.00

0.01

0.32

39.18

195.89

38.79

39.18

13.08

12.96

DRUG 10

5.00

5.00

100

0.0040

500.00

21.31

81.53

500.00

2500.00

76.12

0.017

0.000

0.02

500.00

0.00

0.32

39.36

196.80

38.97

39.36

12.91

12.78

DRUG 11

5.00

5.00

100

0.0100

316.23

18.76

27.53

316.23

632.46

48.14

0.108

0.003

0.11

500.00

0.01

0.13

61.24

122.49

60.66

61.24

8.55

8.47

DRUG 12

5.00

5.00

100

0.0040

500.00

18.70

27.39

500.00

1000.00

76.12

0.043

0.001

0.04

500.00

0.00

0.13

61.95

123.89

61.34

61.95

8.27

8.20

DRUG 13

5.00

2.00

2.9

100

0.0100

316.23

11.99

45.46

316.23

1581.14

48.14

0.043

0.001

0.04

500.00

0.01

0.32

39.18

195.89

38.79

39.18

13.08

12.96

DRUG 14

5.00

2.00

2.9

100

0.0040

500.00

11.98

45.42

500.00

2500.00

76.12

0.017

0.000

0.02

500.00

0.00

0.32

39.36

196.80

38.97

39.36

12.91

12.78

DRUG 15
DRUG 16

5.00
5.00

2.00
2.00

2
2

2.9
2.9

100
100

5
2

0.0100
0.0040

316.23
500.00

11.20
11.18

16.61
16.56

316.23
500.00

632.46
1000.00

48.14
76.12

0.108
0.043

0.003
0.001

0.11
0.04

500.00
500.00

0.01
0.00

0.13
0.13

61.24
61.95

122.49
123.89

60.66
61.34

61.24
61.95

8.55
8.27

8.47
8.20

206.72

204.76

Sum

1112.27

27587.67

Here dummy values are used in sensitivity analysis. The calculations are identical to those used with the actual pharmacy data, which are described in Appendix C.
This analysis was done to demonstrate the influence of the various product characteristics on the determination of the optimal control parameters and order quantities.

110

3085.66

APPENDIX F: MODEL 3 CALCULATION FOR SENSITIVITY ANALYSIS

Tmin

Tmax
3

SL
10

0.99

*M' = Hospital M'

Step 0: Initial Values
M' = Tmax (vidi)

4200.00

M0 = M' - [vi(si-ui)]

3271.80

W = (vidi)
V0 = M02 / 2W2
Iterative Process

78.82
861.49
Iteration 1

Iteration 2

[vi(si-ui)

1101.51

1118.74

Mk = M' - [vi(si-ui)]

3098.49

3081.26
-0.006

Difference

% Change

M0

M1

M2

3271.80
---

3098.49
-5.59%

3081.26
0.56%

Step 0
Drug
DRUG 1
DRUG 2
DRUG 3
DRUG 4
DRUG 5
DRUG 6
DRUG 7
DRUG 8
DRUG 9
DRUG 10
DRUG 11
DRUG 12
DRUG 13
DRUG 14
DRUG 15
DRUG 16

Average
10.00
10.00
10.00
10.00
10.00
10.00
10.00
10.00
5.00
5.00
5.00
5.00
5.00
5.00
5.00
5.00

Sigma
10.00
10.00
10.00
10.00
4.00
4.00
4.00
4.00
5.00
5.00
5.00
5.00
2.00
2.00
2.00
2.00

Size (vi)

ui
5
5
2
2
5
5
2
2
5
5
2
2
5
5
2
2

10.00
10.00
10.00
10.00
5.80
5.80
5.80
5.80
5.00
5.00
5.00
5.00
2.90
2.90
2.90
2.90

si(0)
30.00
30.00
30.00
30.00
30.00
30.00
30.00
30.00
15.00
15.00
15.00
15.00
15.00
15.00
15.00
15.00

Qi

Step 1

(0)

si(1)

58.70
58.70
92.82
92.82
58.70
58.70
92.82
92.82
41.51
41.51
65.63
65.63
41.51
41.51
65.63
65.63

47.08
47.08
41.65
41.65
25.23
25.23
23.61
23.61
21.14
21.14
18.50
18.50
11.93
11.93
11.12
11.12

Qi

(1)

55.59
55.59
87.90
87.90
55.59
55.59
87.90
87.90
39.31
39.31
62.15
62.15
39.31
39.31
62.15
62.15

Step 2
si (2)
47.70
47.70
42.30
42.30
25.42
25.42
23.81
23.81
21.45
21.45
18.81
18.81
12.03
12.03
11.22
11.22

Qi

(2)

55.28
55.28
87.41
87.41
55.28
55.28
87.41
87.41
39.09
39.09
61.81
61.81
39.09
39.09
61.81
61.81

Di
45.28
45.28
77.41
77.41
49.48
49.48
81.61
81.61
34.09
34.09
56.81
56.81
36.19
36.19
58.91
58.91

Step 3
si
47.70
47.70
42.30
42.30
25.42
25.42
23.81
23.81
21.45
21.45
18.81
18.81
12.03
12.03
11.22
11.22

Si
92.99
92.99
119.71
119.71
74.90
74.90
105.42
105.42
55.54
55.54
75.62
75.62
48.22
48.22
70.13
70.13

Here dummy values are used in sensitivity analysis. The calculations are identical to those used with the actual pharmacy data, which are described in Appendix D.
This analysis was done to demonstrate the influence of the various product characteristics on the determination of the optimal control parameters and order quantities.

111

VITA
John Michael Woosley was born in Memphis, Tennessee, to Dwight and Barbara
Woosley; however, his family moved to Baton Rouge, Louisiana, soon after his birth. He
graduated from Redemptorist High School in 1993 and received his Bachelor of Science,
majoring in kinesiology, from Louisiana State University (LSU) in 1997. As an undergraduate,
John was a student-athlete and a 3-time varsity letter winner for the LSU Track and Field
program. While still competing for LSU, John began his graduate studies in the Kinesiology
Department with a concentration in motor behavior. He earned a Master of Science from this
program in 2000.
Upon the completion of his studies in kinesiology, John shifted his focus and started his
second Master of Science program in the Information Systems and Decision Sciences (ISDS)
Department at LSU, which he completed in 2002. In the fall of that same year, he began the
doctoral program in ISDS. While still working on his dissertation, John accepted a position as an
Assistant Coach with the LSU Track and Field program. He held this position for three years
before returning to ISDS to complete his doctoral program.
John has accepted a position at Southeastern Louisiana University in Hammond,
Louisiana. He and his fianc are scheduled to marry in the summer of 2009, and they plan to
stay in Louisiana as they start their lives together. In addition to his academic career, John
continues to work with local high school track and field athletes on a volunteer basis.

112